--- title: "Changelog" description: "Everything we shipped that changes what you can do, newest first, grouped by week." eyebrow: "Changelog" --- # Changelog Everything we ship that changes what you can do, newest first. An entry appears here when the work is merged and live, so this page never describes something you cannot use yet. 359 updates between September 3, 2026 and October 6, 2026. Every one of them is read from a single published list when these docs build, so nothing on this page is typed by hand or left behind. ## Week of October 5, 2026 **October 6, 2026** - **API.** NOLGIA's MCP server can now save and open the document in a connected desktop app, so an agent working in Blender, After Effects, Premiere Pro, Illustrator or Photoshop can save its work without running code. **October 5, 2026** - **Web.** A new guide explains what 24mm, 70mm, 135mm and 200mm do to a frame, how to prompt each one, and how the focal length reel was made, shot by shot. ## Week of September 28, 2026 **October 4, 2026** - **Web.** A new guide shows how to make dreamlike AI videos: one impossible thing in a quiet, tender moment, filmed like a bystander's home footage, with 17 Seedance 2.5 clips and their prompts. - **Web.** Accounts that sign in with a password can now turn on two-step sign-in with an authenticator app or a passkey, and passkeys can sign you in without a password. - **Web.** Share links can now have a password, turn off downloads, or expire after 1, 7 or 30 days or never, and organization admins can turn public links off or require passwords and a maximum lifetime. - **API.** Accounts that sign in with a password can turn on two-step sign-in with an authenticator app or a passkey, and organizations can require it. - **API.** Share links can now have a password, turn off downloads, or never expire, and organization admins can turn public links off or require passwords and a maximum lifetime. **October 3, 2026** - **Web.** The viral nostalgic video guide now shows how to build each shot: hand details, cars, low angles and drone shots, plus keeping two hosts and one look across every shot. **October 2, 2026** - **Web.** The viral nostalgic video guide has a new cover showing both of its projects. - **Web.** The viral nostalgic video guide now also covers action: a man walking away from an explosion, a helicopter over a quiet street and single gunshots in ordinary rooms, with every Seedance 2.5 prompt and the Lyria 3.5 score prompt. - **Web.** A new guide shows how to make the viral home-video clips where one animal that does not belong turns up in an ordinary backyard, with all five Seedance 2.5 prompts ready to copy. **September 30, 2026** - **Web.** Compare Seedance 2.5 with Kling 3.0, find which video model suits each kind of shot, and work out what a clip will cost before you render. Three new blog posts build on renders and scores from tests already published, with current credit prices shown for each model. - **Web.** Every plan now shows what its monthly credits buy, such as about how many Seedance 2.5 videos or GPT Image 2.5 Flare images, at today's prices. - **Web.** Create video now shows the model you picked as a card with a short clip of its own work, and Change opens a gallery of every video model with previews you can play before choosing. - **Web.** GPT-6.1 Sol is now the best brain you can pick for your agent, at 70 credits a turn. It is our strongest reasoning brain, made for long documents and hard work that takes many steps. Agents that ran on GPT-6 Astra now run on GPT-6.1 Sol automatically, at less than half the old rate. **September 29, 2026** - **Web.** You can now connect After Effects, Premiere Pro, Illustrator and Photoshop to NOLGIA. One free extension covers After Effects, Premiere Pro and Illustrator, and Photoshop has its own plugin. Each app has a page with the download, step by step install and troubleshooting. - **Web.** AI apps you connect over OAuth, such as ChatGPT or Claude through our MCP server, now spend your monthly plan credits first and then top-up credits, just like the app. Personal access tokens you create still spend top-up credits only. **September 28, 2026** - **Web.** You can now hear every voice before you pick it, each with its own play button, and create a custom voice from an upload or a recording made in your browser. - **Web.** You can now play a free sample of any voice right beside the Voice picker before you generate speech. ## Week of September 21, 2026 **September 25, 2026** - **Web.** Multicam highlight edits built from only one or two shots now cut between different angles, bringing the original take in as an angle, so a cut never lands on the same angle twice in a row. - **Web.** Signing up with Apple or SSO from a preset or another page now brings you straight back to what you picked, instead of sometimes dropping you on a generic page. **September 24, 2026** - **Web.** Settings → API tokens now shows when each token expires, marks expired tokens so you can revoke them, and flags older tokens that never expire so you know to rotate them. New tokens last 365 days unless you pick a shorter period, from 1 to 365 days. - **Web.** Studio no longer tells you your saved hand edits are waiting for NOLGIA Agent. Edits you make on the timeline show up in the preview and in every export as soon as they're saved, so there's nothing left to apply. - **Web.** Switching speech models in Audio no longer carries over a voice from the previous model. Models without a selectable voice list now show "The model's own voice", so what you hear matches what the form says. - **Web.** Your agent's page no longer says it is being deployed while its chat is already working during an update. - **Web.** Multicam now recognises a product shot: Overhead and Slow push-in come back as real angles, Macro detail is marked Experimental, and angles that need a person in the shot are no longer offered for a product take. - **Web.** The guides on movement now show and recommend Kling v3 and Seedance 2.5, the models that hold up when people dance, sprint or work with their hands, when a drink pours or a car drives by. Every example clip in those guides was reshot on them, and the Veo 3.1 guide now tells you plainly to reach for Kling v3 or Seedance 2.5 when a shot depends on physics, and to keep Veo for spoken dialogue. - **Web.** See your credit estimate update as you choose a video length or poster pack size in Studio, so you can review the cost before submitting your request. - **Web.** Preview all 23 Multicam shots with real video clips before choosing a shot. Hover, focus, or tap the camera-move cards to play a short loop, with captions clarifying the punch-in edit and the slow-arc result shown for dolly zoom. - **Web.** Every preset page now leads with Use this preset, which opens that preset's own composer or its guided chat. A How to use it section lists the steps it will ask you for, and the examples on Multicam and Miroge open with their own files already loaded. - **Web.** Create multicam videos from one screen: preview shot presets, pick individual angles or a ready-made pack, and plan your highlight edit alongside your source video. See the price before you generate, or try an example take to get started. - **Web.** Fine-tune your Studio audio with track and clip volume controls, fade-ins and fade-outs, and voice, music and sound-effect roles. Turn on smart mix to automatically lower background music under speech, and fit music to your timeline with looping and a fade-out. - **Web.** See the maximum credit cost of a guided preset run before you create, with a breakdown tailored to your choices. Pay only for what the run uses, and approve any spending beyond the price shown. - **Web.** You can now tell when a Multicam angle comes back as a copy of your take. See when it is being retried for free, or when it has been refunded, and re-make the angle if you want to try again. - **Web.** The restore and enhance engines are named for what they do everywhere on the site, and Explore's labs no longer include a tile for their vendor; browse them by task (Restore, Enhance) instead. - **Web.** Follow four new step-by-step guides to add voiceovers, music and sound effects to your videos, reuse the same narrator voice across clips, create a faceless narrated channel, and build an ad campaign through MCP. Each guide includes worked examples and screenshots. - **Web.** Learn how to fix common AI video prompt mistakes, make lighting and motion look more natural, reduce hand and object warping, and guide realistic human movement. Four new guides show rendered before-and-after examples and model comparisons, including what did not work. - **API.** The restore and enhance engines are now named for what they do: Proteus Restore, Starlight Restore, Standard Enhance and the rest. Model ids, prices and every preset, job and API call are unchanged. - **Web.** Learn to turn a photo into a video and choose between Seedance 2.5, Veo 3.1, and Hailuo 3 with four new guides featuring copyable prompts, rendered examples, and model limits. - **Web.** Follow four new step-by-step guides to turn a script into a video, create a storyboard and shot list with the NOLGIA Agent, take a short film from script to final cut, and keep characters and locations consistent across shots. Work from complete examples with prompts, settings, and downloadable source media. - **Web.** Follow four new step-by-step guides to create a food or drink video ad, build a car commercial, make ad variations with different products and hooks, and turn a product photo into a video ad and posters. Each guide includes real outputs and practical notes on what worked and what did not. - **Web.** Turn one locked-off take into multiple full-length camera angles with Multicam, all in sync with the original sound. Choose an angle pack or describe your own, get a highlight edit at no extra generation cost, and download every shot or open them together in Studio. - **Web.** After you cancel a render, the chat tile now shows the final credit charge or refund once the provider finishes processing it. - **Web.** Use New world and Swap to edit your own footage with Seedance 2.5 when Gemini Omni editing is unavailable in your region. The model picker defaults to Seedance 2.5 and shows its price before you generate. - **Web.** Compare camera moves from the same still, see where the results miss, and open the composer with a move already selected. Follow a new Miroge guide to swap a product in a clip using a reference photo, see what carries over and what does not, and load the example clip and photo to try it yourself. - **Web.** You can now use Kling v3 and Kling v3 Pro for text-to-video and image-to-video generation, and follow Kling and Seedance model links from guides and blog posts without hitting a missing page. Explore includes every model under a lab and in the Provider filter, making it easier to find models from your preferred makers. Trending labels appear only when fresh usage data is available; otherwise, Explore highlights what is new. - **Web.** Follow the Miroge guide to keep your subject and change the surrounding world in New world mode. Load the skate clip with Use this example, then use the guide to fill in What stays and The new world. - **Web.** A generation you cancel in Miroge now says Canceled and what happened to your credits instead of looking like a failure, and a canceled generation stays in the agent chat after the agent finishes its reply. - **Web.** Find practical guidance on credits and refunds, generating through MCP, choosing prompts, restoring old footage, and fixing clips that came out wrong. Five new guides and six blog posts pair examples with explanations of cancellation, model discovery, timeline exports, and editing existing clips. - **Web.** New Websites and Campaigns pages show what NOLGIA makes for a site and for a launch, with real examples, and the landing page's Everything you ship section links to both. - **Web.** You can cancel a running generation directly from agent chat or the Outputs panel and see it marked Canceled instead of failed or disappearing. Canceled generations also stay clearly labeled on the canvas, so you can tell what stopped at a glance. - **Web.** Miroge examples now have a Use this example button that opens Miroge with the example's video, photos, settings and prompt already loaded, so you can press Generate straight away. **September 23, 2026** - **Web.** A link to Miroge now opens a page that shows what Miroge does, and it shares with its own preview image. - **Web.** Miroge has its own composer. Pick a clip you already have, add a photo of each performer, a pet or a product, and choose Motion transfer, Swap or New world: the moves, the timing and the camera carry over from your clip. The price is on the Generate button and one click runs it. Find Miroge in the Create menu. - **Web.** Generate now starts on one click, with the exact price shown on the button before you press it. - **Web.** You can now cancel running generations from the job queue, progress tray, dashboard, and generation results, or stop a Canvas run. Before you confirm, you’ll see when a refund applies. Canceled generations show a clear status, and clearing finished jobs keeps active generations visible. - **Web.** You can now review clearer explanations of how your data is used, your privacy choices, and the terms that apply to your content, credits, and subscriptions. A new service-provider list and Data Processing Addendum give you more detail about who processes your data and the protections that apply. - **Web.** You can switch to reference images in Veo 3.1 and Veo 3.1 Fast without manually removing your start frame. Veo 3.1 Lite no longer offers unsupported character or reference-image options, helping you avoid failed video requests. - **Web.** You can now download the original model-generated video from the asset page when an original is available, alongside the browser-friendly version. Use Download original to keep the file exactly as the model made it, with codec, bit depth, and file size details shown before you download. - **Web.** Visitors from the EU, the UK and Switzerland are now asked before Google Analytics runs, and anyone can change that choice from Cookie settings in the footer. - **Web.** After deleting your account, you now see confirmation that any paid subscription was cancelled and any NOLGIA Agent was stopped, without implying you had either. - **Web.** You can now choose whether to enable the face likeness check when adding photos to a Character or using a face reference. Turn the check on or off for existing Characters from the Characters page. You can still use your photos and generate images without opting in. - **Web.** You can now delete your account from Settings > Profile. Review what happens to your subscription, agent, and organizations before confirming with your email address, and see the seven-day deletion timeline and how to cancel during the grace period. - **Web.** Get clearer guidance when adding a face reference or casting a character: renders use reference photos and are checked against them, without promising an exact identity match in every shot. - **Web.** Compare six video models on the same prompt with scored results and all 12 clips, then inspect 1080p versus 4K renders with pixel-level comparison sliders. Two new blog guides help you weigh prompt accuracy, image detail, render time and cost before choosing a model and resolution. - **Web.** You can now see which credits the app, CLI and connected assistants use, how agent turns are priced, and when models require a paid plan. Generation notices explain how model providers process your content, and billing settings make clear that auto top-up is unavailable and credits must be topped up manually. - **Web.** Browse the homepage with less video data loading up front: Studio, Agent and Create showcase clips now load as you approach them. When adding a face reference in the image composer, clearer guidance explains that your photo is sent with the render and the result is checked against it, without promising an identical face. - **API.** A generation refused by a model's safety filter is now refunded unless the model provider billed us for the refused attempt, including when the provider's answer does not say either way. The failed job tells you your credits were not charged. - **Web.** You can preview the brand-host and unboxing landing tiles on hover or keyboard focus without loading their videos while idle. Find clearer guidance on the photos and framing that work best for character face checks, and on refunds for content-policy refusals: you are charged only when the model provider bills us. - **Web.** See your audio model’s prompt limit as you type, with a live character count and clear guidance when a prompt is too long. Shorten oversized prompts before generating instead of discovering the limit through a failed generation. - **Web.** Gemini Omni Flash is retired and now shows as Retired in the model picker. Google is shutting the model down on 30 September and names Gemini Omni 1.1 Flash as its replacement, at the same price with the same inputs. The Product commercial preset now runs on Gemini Omni 1.1 Flash, and links to the old model page open the Omni 1.1 Flash page. - **API.** When GPT Image's safety filter blocks a generation, including an image it had already made, your credits now come back. OpenAI does not bill us for a blocked image, so we no longer charge you for one, and the failed job says so. - **Web.** The Guides page has a new design, with a real output at every step, before-and-after comparisons, prompts you can copy and a list of what to watch for, and NOLGIA now has a Blog for launches and model news. - **API.** Gemini Omni Flash is retired. Google is shutting the model down on 30 September and names Gemini Omni 1.1 Flash as its replacement, at the same price with the same inputs. The Product commercial preset now runs on Gemini Omni 1.1 Flash; we compared both models side by side on that preset before moving it. Gemini Omni Flash no longer appears in the model list. From 30 September, a request that names omni-flash is refused with a message pointing to omni-1.1-flash instead of failing at the provider, and no credits are spent. **September 22, 2026** - **API.** Seedance 2.5 no longer fails when you add a reference video and write a prompt that sounds like an edit, such as "Replace the person": the clip is used as a reference at the length and shape you picked. - **Web.** When a model you picked can't use a character's picture, the composer keeps your model, tells you why beside the character controls, and offers the model that can as one tap. - **Web.** The landing page's "Real work. Not demos." wall now features Miroge in place of Outfit try-on. The tile shows Miroge's glacier before/after: a skater's board becomes carved ice and the concrete bowl becomes a glacier, while the rider and the trick stay the same. It links straight to the Miroge preset page. The tile shows a still until you hover it, so it adds no video download to the page load. The presets rail, including its offline fallback, now features Miroge in Outfit try-on's place. - **API.** FLUX Kontext Pro now accepts a reference image, so you can edit a picture with it the way you already could with Kontext Max. - **Web.** The object-swap preset is now called Miroge and lives at nolgia.ai/presets/miroge. Give it a clip and name one object or product to replace, and you get the same clip back with the new object in its place, with the people, the camera, the cuts and the light exactly as you shot them. Older links to the preset still work. - **Web.** Hailuo 3 is not silent. Our copy described MiniMax Hailuo 3 as rendering silent. It renders native 32 kHz stereo audio, including synchronized dialogue, in the same pass as the picture (shaped by the prompt's "Sound:" clause). The customer-facing audio chip was already correct (it renders "native audio" from live /models); the stale claim lived in developer copy that could have driven it wrong again. Video-to-video now defaults to Seedance 2.5. "Add VFX to footage" opens on Seedance 2.5, which takes up to ten reference videos of 2-30 s each (plus reference audio), instead of Seedance 2.0 Pro's three at 2-15 s. The Gaia restore engines were labelled backwards in the restore picker. Gaia High Quality is the live-action engine and Gaia CG is the animation / 3D one; the descriptors were swapped, so anyone restoring animation was steered to the live-action engine. - **API.** The Gaia restore engine descriptions were inverted. topaz-gaia (engine ghq-5) is the live-action Gaia High Quality engine and topaz-gaia-cg (gcg-5) is the animation / 3D-render one. The model list had the two roles swapped, so anyone restoring animation was pointed at the live-action engine and anyone restoring live action at the CG engine. Both lines now read correctly everywhere the list is published (API reference, SDKs, MCP). minimax-h3 no longer advertises reference videos. Hailuo 3 accepts none (video_refs_max: 0), the MiniMax provider refuses video_urls. The model card was already correct, but the MCP tool descriptions still told an agent it took up to 3, so agents were being steered to send input the provider rejects. The video_asset_ids and prompt-token descriptions now say minimax-h3 takes no reference videos. Its 9 element images and 3 reference audio tracks are unchanged. - **API.** Starlight, Starlight Fast and Starlight Sharp restores work again. They were failing every render. Artemis Moire restores work again. It was pointed at an engine the vendor had stopped processing. - **API.** Artemis Moire restores work again. The engine we were routing to had started failing every job server-side; this points the model at the working one. Same quality target, same price. - **API.** The object-swap preset is now called Miroge and lives at /presets/miroge. Give it a clip and name one object or product to replace, and you get the same clip back with the new object in its place, with the people, the camera, the cuts and the light exactly as you shot them. Links and scripts using the old recast-swap id keep working. - **API.** The API documentation now calls the agent product by its name. NOLGIA Agent was described throughout the public API spec and developer guides by an internal codename, which made the docs for a product you can buy read as though they were about something else. Every description, endpoint summary and guide page now says NOLGIA Agent. Nothing about the API itself changed: the X-NOLGIA-Surface header still accepts exactly the same values it always has, and existing integrations are unaffected. - **Web.** The billing page now explains which credits API access actually spends. It previously said extra credits are used after your plan credits run out, which is the wrong way round for anyone calling the API with a personal access token: those calls spend extra credits only and never touch your monthly plan credits. If you have a plan balance and an API call was refused, that is why. The billing page now says this in the same words as the pricing page, including when you have a monthly grant. Nothing about how credits are spent has changed. - **Web.** The public changelog no longer shows internal engineering notes on entries whose authors had marked them as having no customer-facing change, and the NOLGIA name is spelled consistently across every entry. - **Web.** Video models in Explore now show a preview the model made itself, instead of a name plate. Each card shows a still at rest and plays a short clip when you hover, so browsing the catalog stays fast and nothing downloads until you ask for it. **September 21, 2026** - **API.** A fault in our catalog self-check has been corrected: two talking-avatar models were being reported as broken when in fact the check was not sending them the voice track they need. They render normally and were never affected, only the internal report was wrong. - **API.** The daily check that proves every model in our catalog still works now asks video models every other week instead of every week, which cuts what the check costs by about a third. Image, audio and 3D models are still checked weekly, and every model is still checked on a fixed, published schedule, no model goes unexamined for longer than a fortnight, and one that stops working is still flagged within two weeks of its first failure. - **Web.** Most models in Explore now show a preview made by that model, chosen to show what it is actually best at, so an image model shows an image and no two cards share a picture. Most model pages now start with a real example: the prompt that is already in the box, and the result that prompt produced, so you can see what a model does before you spend a credit. - **Web.** Model ids are now ours rather than our suppliers' route paths, so every model has a reference page, including all 17 audio models, which had none. Old ids keep working permanently, and old model-page links redirect. - **API.** Every model in the catalog now has an id that is ours: elevenlabs-tts-v3 instead of fal-ai/elevenlabs/tts/eleven-v3, seedance-2.0-pro-i2v instead of fal-ai/bytedance/seedance/v2/pro/image-to-video, flux-schnell instead of flux/schnell. Twenty-seven ids change, seventeen of them audio. The old spellings keep working, permanently, not on a deprecation clock, so nothing you have already written needs to change: a script, a saved preset or a stored job that names the old id resolves exactly as it did, and it is only the name we publish that moves. Those ids named a routing detail that was already out of date, which is the better reason to retire them than tidiness: twelve of the twenty-seven named a supplier we do not buy them through at all. The models that carried them also now have documentation. The reference generator skips an id containing a slash, so those twenty-seven had no page and the audio category had none at all; the reference covers all 143 models today, with seventeen audio pages where there were zero. - **API.** Lyria music generation now runs on NOLGIA's own Google credentials instead of a reseller, on Google's current flagship Lyria 3.5. The model id, the audio format and the price are unchanged. A superseded veo-3 route that no customer could select has been removed; veo-3.1, veo-3.1-fast and veo-3.1-lite are unaffected. - **API.** NOLGIA now checks its own model catalog. Every day a rotating slice of the models we offer is asked to produce something, so that within a week every listed model has been proven to still work, and anything that has quietly stopped working is flagged for us rather than discovered by a customer spending credits on it. - **API.** Seedance 2.0 Pro reference-to-video renders at 720p now run on a cheaper backend. Output, capabilities, limits and credit cost are unchanged at every resolution: the same reference videos and element images are accepted, and a clip costs the same number of credits as before. Renders at 1080p and 4K are unaffected. - **API.** Lyria music generation now runs on NOLGIA's own Google credentials instead of a reseller, on Google's current flagship Lyria 3.5. The model id, the audio format and the price are unchanged. - **API.** Lyria music generations now meter at their real $0.08-per-song rate instead of recording $0.00, and audio returned through the speech bridge is labelled with the container it actually is rather than always audio/mpeg. Nothing changes for callers of models that were already correct. - **Web.** Reve 2.1 is no longer selectable and appears as Retired. Reve is shutting down: its API stopped serving on 14 August and the service ends entirely on 27 September, so every Reve generation was failing. Existing presets, saved settings and links that reference Reve still open rather than breaking, and no further credits will be spent on a model that cannot produce a result. - **API.** Reve 2.1 has been withdrawn from the model catalog. Reve is shutting down: its API stopped serving on 14 August and the service ends entirely on 27 September, so every Reve generation was failing. Saved presets and scripts that name reve-2.1 will keep resolving rather than returning an unknown-model error, and no further credits will be spent on a model that cannot produce a result. Every other model is unaffected. - **Web.** Selecting assets in your Library now offers a Download. A few land in your Downloads folder as themselves, named the way they are named on screen; a larger selection arrives as one.zip with a progress count while it builds. Mixed batches work, so images, video, audio and 3D all come back with the right extension and open in the app you expect, and anything that could not be included is named rather than quietly missing. - **Web.** Exporting from the timeline is one button now. Press Export video to render your cut as before, or open the menu beside it for everything else in one place: a PNG still of the frame at the playhead, with or without a transparent background, and project files for Premiere Pro, DaVinci Resolve or After Effects. No more guessing which of three Export buttons hid the one you wanted. - **API.** The documentation site now has a section for Organizations and a proper Agent branch. Team administration (members, invites, verified domains, organization API keys and bring-your-own storage) has its own page instead of being buried in one long page, and the NOLGIA Agent now has an introduction and a page that says plainly which plans can run one and what a turn costs. - **Web.** The developers page now shows one install command per client, npm install @nolgia/sdk, pip install nolgia, cargo add nolgia-client, because each package brings everything it needs. Rust also gets its own card: until now the page offered only the nolgia command-line binary, with no way to find the Rust library. - **API.** The documentation site now has a Changelog page. Everything we ship that changes what you can do is listed there, newest first, grouped by week, with the parts of NOLGIA each update touched. It is rebuilt from a single published list every time the docs publish, so it cannot fall behind what actually shipped. - **Web.** Every model now has a playground: open a model page, fill in a generated form that shows only the settings that model actually accepts, and see the exact price of those settings on the button before you spend anything. An API tab beside it shows the same request as code in TypeScript, Python, Rust, the CLI and curl. The whole page works before you sign in. - **Web.** Everything we ship is now published as a plain list your own tools can read, at nolgia.ai/changelog/entries.json. Each update carries the day it shipped, what it lets you do and the parts of NOLGIA it touched, so an agent or a script can follow what changed without reading the page. - **API.** Models that turn a photo and a voice track into a talking head now say so on the public model list, so any app or page can offer the right controls for them without guessing. Several common settings, including aspect ratio, audio format, negative prompt, seed and the number of images, now carry a plain explanation everywhere they appear. - **API.** Every install tab in the docs is now a single command. npm install @nolgia/sdk, pip install nolgia, cargo add nolgia-client, each package brings everything the examples need, and the note explaining extra packages is gone because there are none. The Rust quickstart saves its image and video with the client's own download, so it no longer asks you to add an HTTP crate. - **API.** The docs now have a single API reference entry point that tells you which of the five ways to call NOLGIA, REST, TypeScript, Python, the Rust CLI or the MCP server, fits what you are doing, with the current versions and server URL on the page. - **API.** Every model on NOLGIA now has a reference page showing who made it, what it costs, what it can do and a call you can copy and run. Browse them by what you want to make, image generation, image editing, text to video, image to video, audio and speech, restore and upscale, utilities, or 3D. - **Web.** The Go client is not installable from a public module path today, so it is no longer advertised on the developer page or in the quickstarts. Rather than print an install command that cannot succeed, we removed it until the Go module is published publicly. Python, TypeScript and the Rust CLI are unaffected. - **Web.** A brand new account's dashboard now suggests three presets you can actually run. - **Web.** Your dashboard now opens with your own most recent work: the last six images, clips and cuts you made, big enough to actually see, one click from the full file. - **Web.** Explore is live. Browse every model and preset on NOLGIA by task, by lab, or by what is trending this week, and search across all of them with results that update as you type, with no Enter and no page reload. Every public model now has a page, including the Kling tiers, the 3D engines and the restore, upscale and background removal tools that were not listed before. - **Web.** Getting around the app is simpler. Every place now has one name: the page you land on after signing in is "Dashboard" everywhere instead of being called three different things, and the chat is "Agent" everywhere instead of "Create video" in one place. The top bar no longer repeats what is already in the sidebar, so it holds just Create, Agent and Timeline beside your account. - **API.** npm install @nolgia/sdk is now the complete install for the TypeScript SDK, no second package to add, on any package manager. - **API.** The @nolgia/sdk (npm) and nolgia (PyPI) packages now ship subscribe and submit: submit a generation and get the finished media URLs back with no hand-written polling loop, with typed NolgiaGenerationError codes when a generation is refused or fails. - **Web.** The app fits a phone screen. Timelines, Projects, a project, an asset, the image and audio consoles and Characters no longer scroll sideways, the Agent's message box is a real box again instead of a sliver between buttons, every page names itself in the browser tab and in your history, and the Studio preview shows your text layers in the fonts you picked. - **API.** docs.nolgia.ai has a three-column layout: a collapsible navigation, the page, and a table of contents that follows you, with a drawer on phones. - **Web.** Every link inside the app now opens the page it names. Project titles open the project workspace directly, "Use in Studio" and "Open in Studio" open the timeline editor on that clip, "All timelines" and "Open Studio" open the Timelines page, and "Reuse in generate" opens the image, video, audio or 3D console with the model preselected, with no detour through a redirect. ## Week of September 14, 2026 **September 20, 2026** - **Web.** The site has one navigation everywhere. The header now reads Presets, Product, Developers, Teams and Pricing on every page, with a menu on phones that reaches every section, and the footer groups everything into Make, Work with it, Plans and Company. If you are signed in, the header offers Open app and Create video instead of Sign in. Links on the home page all open live pages again, and every card in "Everything you ship" opens the tool that makes it. - **API.** Video renders settle within seconds of finishing on providers that call back. Kling is first: the callback wakes the poller immediately and the poller only reconciles, measured at three status reads per render instead of fifteen. - **Web.** The CLI and MCP pages list the live tool surface, code blocks across the site are highlighted with a copy button, and a new visual kit covers the marketing pages. - **API.** docs.nolgia.ai has nine spec-verified guides: quickstart, authentication, jobs, uploads, errors, billing, organizations, the Agent Sessions API and client libraries, each also served as Markdown for agents, with llms.txt. - **API.** Kling 3.0 Turbo, Kling 3.0 Omni and Kling O1 are in the catalog. - **API.** Background removal as two zero-prompt utilities, for images and for video. - **Web.** Paste this into your agent. One line on the CLI, MCP and developer pages that has your coding agent install the CLI, sign you in, install the skills and run a verification generation. **September 19, 2026** - **Web, Billing.** The real server price before a real spend. Every generation shows the exact credits it will hold, from the same calculation that bills it. - **API, Billing.** Know the price before you submit. POST /jobs/cost returns the exact credits a generation will hold and a confirmation token you send back with the request; every failure carries a typed error code, and a duplicate submission answers 409 naming the job that already exists. - **Web, Editor.** Export the current Studio frame as a PNG, rasterized on the server with the 22 Studio typefaces. - **API.** Kling 2.6, Kling 2.5 Turbo and Kling Avatar are in the catalog, and the 4K audio surcharge is gone. - **Web, API.** Surgical tweaks. Pick a measured tweak scope such as expression or background in the edit strip so the model changes only that. - **API.** Kling motion control. Transfer a human performance from a reference video onto a new subject. - **Web, API.** Style keys. A saved style with a swatch that rides on any video generation, and a style slot on each project that every clip inherits. - **Web.** Preset cards share one fixed-ratio media frame, so a vertical clip and a wide image both fit. - **Web, API.** Sets. Generate a coordinated pack or a set of variants as one request; each member is its own job, shown together in Outputs and on the project, with Make variants on the composer. - **Agent.** UGC presets cast a creator of the age and demographic you describe by default. - **Web, Editor.** Studio text, second edition. Stroke, paint order, spacing, caps, shadow presets and five headline looks, rendered identically in the export. - **Agent.** Commerce presets, with a main-image check that verifies a listing image meets marketplace rules before you use it. - **API.** Speaking rate and captions. Set a speed on text to speech, and burn timed captions into a narrated blocks render. - **Agent.** Thumbnail presets add a logo step, a style brief and a feed-size proof that checks the result at the size a feed shows it. - **Agent.** Brand mockups. A preset family that places your brand on real-world objects, drawing on the brand kit. - **Agent.** Restyle. Re-skin an existing image with a seasonal or aesthetic look while keeping its subject. - **Agent.** Blender on your Agent. A headless Blender 4.5 LTS ability runs on the Agent pod for 3D work. - **Web.** Presets assemble their prompt on the server from your answers. Preview the assembled prompt on a preset run, keep it or revert. - **Web, Editor.** Import a Figma frame into a Studio timeline. - **Web, Editor.** Export a Studio timeline for Premiere Pro, DaVinci Resolve or After Effects, with the colour grades as LUTs and text as native layers where the format allows. - **Web, API.** 3D from photos. Turn one to four product photos into a GLB model, view it in the browser, and make one through the API or MCP as well. - **Web, API.** Keep editing. Chain image edits step by step, revert to any earlier step, and save a style you can apply again. - **Web.** Arena. Public creative challenges you enter with a finished Output, with community voting. - **Web, Agent.** A node canvas per project. A production plan that you and the Agent build together, versioned. - **Web, API.** A voice on every character. Narration, dialogue and voice-reference video speak with it, and the audio composer can Speak as a character. - **Web, API.** Transcripts and shorts. Transcribe any video or audio in your Library and find short clips inside a long video. - **Web.** A generation blocked by a provider's content filter now says so on the card, tells you whether the attempt used credits, and offers Edit prompt to put your prompt back in the composer. - **API, Billing.** Refused generations are refunded. When a provider's content filter blocks a render, the job says so with a typed outcome, and the credits come back whenever the provider did not bill the attempt. - **Web.** The developer portal carries guides, every CLI install path, the skills and plugins, and the Agent Sessions API. - **Web, API.** Recast. Swap one object in a clip with Gemini Omni 1.1 edit mode. - **API, Agent.** Narrated explainers. Ordered clip-and-narration blocks assemble into one MP4 through POST /renders/blocks. - **Web.** Public pages for every model at /models with capabilities and rates, and a public Characters page. - **Agent.** Video scorecard. Ask the Agent for a Hook Score and a scorecard on a finished clip. - **Web.** A guides hub at /guides with preset guides, launch posts and an Agent cookbook. - **Web, API.** Import a product from a store link and reuse it in every ad; product intake steps in the composers and a Products section on each project. - **Web.** Characters show whether their references are ready, with a photo guide and per-character preferred models that the composers honour. You can take reference photos with your camera. - **API, Agent.** The Agent Sessions API. Create persistent sessions, send messages, stream replies, interrupt a running turn, answer the Agent's questions, and branch on typed error codes; a Stop button in the chat and in Studio. - **Web, API.** Brand kits. Keep palettes, fonts, logos and never-rules together and apply them to a project or a single generation. - **Web.** Share pages offer Make your own, Studio has a public page, and pricing pages say what things cost. - **Web, API.** Camera moves. Pick a push-in, orbit, crane, rack focus or any move from the library, with a strength, for any video generation. - **Web, API.** Cast up to four characters in one generation, reordered with chips; each member gets its own identity score on the result. - **API.** Seedance 2.5 does more from one model id. It renders 1080p, takes up to ten reference videos and ten audio tracks, and runs three tasks: reference a clip, edit it, or extend it. - **Web, API.** Locations. Save a place with a description and reference images, and set any generation in it. - **Web, API.** Reframe an image to 9:16, 1:1 or 4:5. The new margins are painted in and the original pixels stay untouched. - **Web.** A public page for the NOLGIA Agent, with quickstarts that match the SDK READMEs. - **Web.** Product photo and marketplace listing presets, plus a stock presenter picker for the UGC presets. **September 18, 2026** - **Web.** Share links count views. See them in the share sheet and manage every link you have made from Settings, Share links. - **Web.** Extend a video. Continue after its last frame or lead into its first, from the asset page. - **Web.** Characters carry motion reference clips and wardrobe outfits, picked from the character panel when you generate. - **Web.** See the prompt that actually rendered next to your own on every image, and enhance a video prompt for the chosen model with one click. - **Web.** Projects can form a series. Start the next episode from the current project; its cast of characters is attached automatically in every composer and Agent turn. **September 14, 2026** - **Web.** Notify me when done. Turn on browser notifications for Agent turns, generations and exports and get told the moment they finish. - **Web, Agent.** Bring your Library into a chat. Attach existing assets to an Agent message, and the Create button always acts on what you have said so far. - **Web, Editor.** Studio opens on a Timelines page for each project. In the editor, drag to marquee-select clips and press Cmd+B to split at the playhead. - **Web, Agent.** Start a chat from a preset. New chats open on a preset grid with visual starter cards; the chosen preset stays pinned at the top of the conversation. - **Web.** When a generation fails, the reason is shown wherever that generation appears, and the activity feed carries it too. ## Week of September 7, 2026 **September 13, 2026** - **Web, API.** Generations run in parallel by plan. Free accounts run one generation at a time, Starter two, Pro four, Studio and Team eight, shared across image, audio and video. Your live count shows on the account page and on GET /me. - **API.** Two new video lanes. HappyHorse reference-to-video renders from one to nine reference images with audio, and HappyHorse video edit rewrites a clip you supply from a text instruction. **September 11, 2026** - **Web.** Fixes guided preset loading, the lucky starter shortcut, skipped inputs, and recovery after direct generation. - **Web.** You can now create an account with an email and password, alongside Google and Apple. - **Agent.** NOLGIA is now packaged for the OpenAI Plugins directory. A new plugins/chatgpt/ directory carries the manifest, six customer-facing skills and the full submission checklist, pointed at the MCP server we already run at https://mcp.nolgia.ai/mcp. - **Web.** Starting a preset now opens a guided chat that asks for your material step by step. - **Web.** Presets that guide you step by step now carry a Guided tag. **September 10, 2026** - **Web.** Presets now guide you step by step through the inputs they need, from your own logo, photos, footage, audio or data. - **API.** Presets can now describe the inputs they need, so the app can guide you through them. - **Agent.** Characters in chat are shown only for presets that use them, with a searchable panel instead of an ever-growing list. - **API.** Animated cartoon is now a family: pick a style branch (storybook, anime, Saturday morning TV, retro rubber hose, stop motion clay, paper cutout, 3D feature look, preschool) for a one-tap scene, or Full episode for a whole episode with the same characters, a title and a score. - **API.** Preset suggestions have time to use their expanded reasoning budget and only salvage incomplete replies when the provider reports token-limit truncation. - **Web.** Anyone can browse the preset catalog and every preset page without signing in. - **Web.** The agent page now streams progress instead of polling, so it uses a fraction of the requests - **API.** Motion graphics now branches into eight directions, each with its own example - **Web.** Empty chats are no longer saved to your history: a new chat is created only when you send its first message. - **API.** Media loads faster: preview links stay stable so your browser can cache them, and videos start playing before they finish downloading. - **API.** Preset suggestions give the model room to think, so a brief now gets its directions instead of a timeout. - **API.** Preset suggestions stop retries in time to refund a failed request before the gateway deadline. - **Web.** The app makes far fewer background requests: pages load once, polling pauses when a tab is hidden, and static files are cached. - **API.** Preset suggestions wait long enough for the model to finish writing. - **Web.** Preset pages now offer directions: a parent preset shows the versions under it plus a "describe your own" path, and every preset page can suggest directions tailored to your brief for 1 credit. - **API.** Directions under Musical performance run as one-tap videos like their parent. - **Web.** Fixed a bug that made the app re-check your session constantly in the background. - **Agent.** Your NOLGIA Agent can now edit part of an image with a mask, add lip sync from a voice track, and pick how much detail the GPT Image models spend on a render. - **API.** Preset pages can now offer directions under a family, with suggestions tailored to your brief. - **API.** feat(presets): generalize the logo and musician presets and rename two cards (migration 000102) - **Agent.** The logo animation playbook now covers any original studio-style opener, and the musical performance playbook covers any instrument, performer and setting. - **Web.** Twenty-four new presets, each published with three examples. - **Web.** Preset pages can now show audio examples. - **API.** Preset pages can now show audio examples. - **Web.** The region editor now shows how much of the image you have painted and warns when the window is too small to hold the object or so large the model tends to return a blank region, and both the painter and the request summary explain that the prompt should describe the whole scene, not only the change. - **Agent.** presets: spreadsheet and animated spreadsheet, measured (preset-workflows 1.14.18) - **API.** Editing part of an image with a mask is now offered only on the models that keep the rest of the picture intact, and the API explains how to write a prompt the mask will follow. - **API.** Microsoft MAI-Image 2.6, 2.6 Flash, 2.5 and 2.5 Pro now edit: one reference image per request, downscaled to a 2048 px long edge before dispatch and published as image.reference_long_edge_max. Their credits move to cover that render: 6, 3, 6 and 10 at native, with the 2k/4k upscale tiers up by the same two credits. - **API.** MiniMax H3 Max runs on the direct provider and takes references on its text-to-video id. - **Web.** GPT Image models now have a Render quality setting in the image console, separate from the upscale tier: Auto, Low, Medium and High cost the base rate, while Extra high and Max on GPT Image 2.5 add 11 and 24 credits per image, shown before you generate and on the pricing page. - **API, Billing.** GPT Image models gain higher quality settings: extra high and max. - **API.** Video presets keep their sound on by default, and image prompts about objects no longer get portrait styling added. - **Web.** Six new video models are live, led by Wan 3.0 as the new text to video default, and Wan 2.7 brings reference image control to the Wan family for the first time. - **Web.** Edit just one part of an image: on the GPT Image models, attach a reference, paint over the region to change, and only that area is regenerated. - **Web.** Show the seven OpenRouter image models, and never offer them a reference - **Web.** A new Enhance Image tool makes a picture from your library bigger and cleaner at up to 4K, keeping the same composition, for 7 credits. - **API.** Five new image models enhance a picture you already have: Standard, High Fidelity, Low Resolution, Art & CGI, and Text & Shapes, at up to 4K for 7 credits. - **Agent.** Agents now enhance an image you already have instead of re-rolling the prompt when you ask for a bigger or sharper version. - **Agent.** Bumps preset-workflows to 1.14.15 so the video roster doctrine merged in actually reaches the marketplace package. - **Agent.** Six video routes join the model matrix and every agent-facing roster, with wan-3.0 documented as the default text-to-video model and wan-2.7 as the reference-to-video lane for real people. - **Agent.** The ten Astra cards are live as private drafts, and the playbooks record what three production re-rolls taught about sets, signage and promises. - **Web.** The video composer can now lip sync a portrait to a voice track, and restyle footage you already have from a written instruction. - **API.** Image enhancement is now available: five precision upscalers that sharpen and enlarge an image you already have, at 4K output. - **API.** Lip sync and video editing are now available through the API, so a portrait and a voice track become a talking head, and footage you already have can be restyled from a text instruction. - **Web.** The voiceover console now shows what your script will cost as you type it, because voices are billed by the length of the text. - **API, Billing.** Fewer wasted renders: repeated requests with rejected references are stopped early, and you are only charged when the provider charged us. - **API.** Wan 3.0 is available again and is now the default text-to-video model, joined by Wan 3.0 Prime, Wan 2.7 (reference-to-video), Wan 2.6, HappyHorse 1.0 and Seedance 1.5 Pro. - **API.** Adds seven image models to the catalog: Microsoft MAI-Image 2.6, 2.6 Flash, 2.5 and 2.5 Pro, Sourceful Riverflow V2.5 Pro and Fast, and ByteDance Seedream 4.5. - **Agent.** The platform skill now tells agents that voiceover is billed by the length of the script, so they stop batching lines to save a generation. - **API, Billing.** Voiceover now costs by the length of your text: short lines cost less, long scripts cover their cost. - **Web.** The Restore console now groups engines by what they fix, compression damage, interlacing, grain, faces, and says in plain words what each one is for. - **API.** Lip sync and video editing. HeyGen Avatar IV turns a portrait plus a voice track into a talking head, and Runway Aleph 2 rewrites footage you supply from a text instruction. - **API.** Edit part of an image: paint a mask and the GPT Image models change only that area. - **Web.** Connecting an assistant to NOLGIA over MCP now signs you in through NOLGIA instead of asking you to paste a token, and the CLI and MCP pages publish the same, correct server address. - **Agent.** Ten new preset playbooks completing the GPT-6 Astra showcase, and every preset in both rounds renamed after the card it came from. - **API.** Connect from any MCP client. mcp.nolgia.ai supports OAuth 2.1 sign-in for connectors, with refresh tokens so a connector recovers without you, and tool annotations that keep destructive tools marked as such. - **API.** Seventeen more restore engines are now available, grouped by what they fix: denoise, deinterlace, up-res, face recovery, compression repair, motion deblur and SDR to HDR. - **API, Billing.** Agent chat turns are now billed on their actual cost, never less than the published per-turn rate. - **Agent.** Model doctrine now quotes live credit rates instead of hard-coded ones, and the price ladders that remain (the video/image roster and the restore engines) are re-priced against the published price list and dated. - **Web.** The agent brain picker offers GLM 5.3 in place of GLM 5.2. Agents still pinned to GLM 5.2 or Kimi K3 can review the retirement notice and explicitly choose a replacement. - **API.** Nine new image models are now available: Microsoft's MAI-Image 2.6, 2.6 Flash, 2.5 and 2.5 Pro, Sourceful Riverflow V2.5 Pro and Fast, Qwen Image 3 Pro, and Seedream 5.0 Lite and 4.5. - **API.** Five more video models are now routable: Wan 3.0 Prime, Wan 2.7, Wan 2.6, HappyHorse 1.0 and Seedance 1.5 Pro. MiniMax H3 Max stays on its existing fal routes until its API contract migration is deployed. - **API.** Twenty-one more restoration engines are now routed through the video proxy. - **API, Billing.** Grok Imagine video and image prices now reflect each resolution's real cost. - **API.** Presets that used Kling now run on Seedance or MiniMax H3. - **Agent.** Model doctrine now quotes live credit rates instead of hard-coded ones, and the two price ladders that remain are re-priced against the published price list and dated. - **Agent.** Twelve new preset playbooks distilled from the GPT-6 Astra showcase, each with its element checklist registered so its first run is gated. - **API.** Agent brains: GLM 5.3 replaces GLM 5.2, and the roster is now four brains. - **Agent.** Agent brains. GLM 5.3 replaces GLM 5.2, Kimi K3 is retired, and GPT-6 Astra is available as a brain. - **Agent.** Presets that used Kling now run on Seedance or MiniMax H3. - **Web.** New NOLGIA MCP page: connect NOLGIA to Claude Code or Cursor and recreate any preset from inside your assistant. - **Web.** Presets that used Kling now run on Seedance or MiniMax H3. - **Agent.** Presets that used Kling now run on Seedance or MiniMax H3. - **API.** Generations no longer fail when our servers scale down mid-render; interrupted work resumes automatically. - **Web.** Preset cards are cleaner: click a preset's name to open its page; the separate Details button is gone. - **API, Billing.** Prices for some video and audio models rose by about 10 percent so every model covers its cost at every plan. - **Web.** GPT-6 Astra can now be chosen as your agent's brain. - **Agent.** Your NOLGIA Agent runs the latest command line tools, so it can read each model's prompt limits before it spends anything and work with share links, your credit history and preset details directly. - **API.** _No user-facing change._ - **Web.** Kling models are no longer selectable while the provider account is unavailable; the pickers show them greyed out and Create opens on its usual default instead. **September 9, 2026** - **API.** Image presets now only use brands you mention and never add real logos on their own. - **API.** Kling models are temporarily unavailable while we restore the provider account. - **Web.** FLUX 3 Video is listed in the video picker and on the pricing page. - **Agent.** Your agent can now run GPT-6 Astra, OpenAI's most capable model, for the hardest work: complex reasoning, coding, research and long documents. It costs 170 credits a turn against the default brain's 10, it takes effect on your very next message, and your agent will quote you the price and wait for you to ask before it ever moves you onto it. - **Agent.** The agent no longer suggests GPT Image 2 for new work and reaches for GPT Image 2.5 instead. - **API.** Sell GPT-6 Astra as an agent brain at 280 credits per turn - **Web.** GPT Image 2 is marked legacy and moved down the image picker; GPT Image 2.5 is the recommended choice. - **API.** Text to speech requests longer than 4,000 characters are now refused with a clear error instead of being sent to the provider. - **API.** Attaching reference images that are too large together now fails instantly with a clear message, instead of starting a render that dies a few seconds later. - **API, Billing.** Kling Image 3.0 now costs 3 credits instead of 112, and ElevenLabs Music 60 instead of 70. - **Web.** Switching your agent's brain now takes effect on the next message. - **API, Billing.** Corrected the credit price of two image models that were being sold below their provider cost: GPT Image 1 now costs 13 credits (was 4) and Grok Imagine Image 2.0 now costs 3 credits (was 2). **September 8, 2026** - **Web.** The pricing FAQ now quotes the real range of video clip prices from the live price list. - **Web, Billing.** Reference images without a reference video now show the correct reference-to-video price in the video composer, instead of a lower estimate and a note about a discount that no longer applies. - **Agent.** Preset doctrine now quotes the live video prices: MiniMax H3 768p drafts at 33 credits per 5s, Seedance 2.0 reference-to-video at 76, and flux-3-video continuation at 108/135 when a reference video is attached. - **API, Billing.** Credit prices for Seedance 2.0 reference-to-video, Hailuo 3 at 768p and Flux 3 Video continuation renders now reflect their measured provider cost. - **Web.** Preset names across the site now match their plain-language names (for example Social media ad instead of UGC Ad). - **Agent.** Preset production workflows now print the same plain-language preset names customers see on the cards, with each preset's slug kept in the title so routing and delivery gates stay addressed by slug. - **API.** Timeline renders now appear in your usage history. - **API.** Five image presets now run on GPT Image 2.5 Flare instead of GPT Image 2: Billboard, Cover art, Logo, Poster and YouTube thumbnail. Each was validated with a real render on the new model first, using the preset's own production prompt. Flare is faster and, with the GPT Image 2.5 re-pricing, cheaper to run. - **API.** Presets now say what they are in plain language. The catalog no longer uses acronyms or insider shorthand in a preset's name ("UGC Ad" is now "Social media ad", "Fit Check" is "Outfit try-on", "App and SaaS Ad" is "App ad"), and every name is written in sentence case. Links, saved work and the preset pages are unchanged: only the display names moved. - **Web, Billing.** GPT Image 2.5 Sunburst and Flare now cost 8 credits per image, down from 22, after measuring their real cost. - **Agent.** GPT Image 2.5 Sunburst and Flare now cost 8 credits instead of 22, making them both better and cheaper than GPT Image 2, and the agent fleet's doctrine has been corrected to match. - **Web.** The API pricing page now shows every model's credit rate as soon as you open it, and keeps up when a price changes. - **API.** GPT Image 2.5 Sunburst and Flare now cost 8 credits instead of 22 (2K 21, 4K 58), after measuring that they use far fewer tokens per render than GPT Image 2. - **Web.** Marketing links to a preset now carry their campaign into the chat they start, so campaigns can be measured. - **API, Agent.** Switching your agent's brain now takes effect on your next message instead of waiting several minutes for the agent to restart. - **Web.** The API pricing page lists published per-model credit rates by quality tier, keeps pending prices visible, and includes every selectable audio model. - **Web, Editor.** Keyframes can now be copied and pasted between clips, and every keyframe carries an easing picker with a live preview curve. - **Web.** New image generations default to GPT Image 2.5 Flare; Sunburst stays one click away for detailed work. - **API.** Chat sessions launched from a preset can now record the marketing campaign that produced them. POST /agent/sessions takes an optional campaign object carrying the five standard UTM dimensions, the session read model serves it back, and a new admin report counts launches per preset per campaign over a date range. - **Agent.** The agent fleet's image doctrine is back on GPT Image 2 as the default, with the GPT Image 2.5 pair explicitly gated as not yet servable while OpenAI organization verification is pending. - **API.** Model prices are now published at GET /pricing/models, readable without an API key, so a price list can be built from the same numbers the API bills from. - **Web.** New image generations default to GPT Image 2.5 Flare; Sunburst stays one click away for detailed work. - **Web.** GPT Image 2.5 Sunburst and Flare are available in the image picker, with Sunburst as the new default. - **Agent.** GPT Image 2.5 Sunburst and Flare are available in the image picker, with Sunburst as the new default. - **API.** GPT Image 2.5 Sunburst and Flare are available for image generation and editing. - **API.** GPT Image 2.5 Sunburst and Flare, with two quality axes: how much detail the model spends drawing, and the upscale ladder. - **Web, Billing.** Per-credit prices on the pricing page and the top-up control now read in dollars with their exact decimals instead of cents. - **Web.** The pricing page now has an Individual / Business & Enterprise / API switch, per-audience plan cards with live team seat math, a searchable feature comparison, and a grouped FAQ. - **Web.** A public What's new page at nolgia.com/changelog lists product updates as they ship, grouped by week. - **Editor.** Keyframes are easier to discover: click an empty lane to add one, use the plus button at the playhead, and no Alt click is needed. - **Editor.** Motion properties can be keyframed over time, with expanded track lanes for editing keyframes. - **Billing.** Billing shows your usage history with a chart, an itemized ledger and CSV export. **September 7, 2026** - **Web.** Preset pages show use cases, examples, longer descriptions and the recommended models. - **Web.** Pricing is easy to find from the landing page on every screen size. - **Billing.** The billing page has one clear layout, and you can change plans in the app with the cost shown before you confirm. - **API, Billing.** The API returns an itemized credit ledger and a usage summary for your billing period. - **Web.** Preset pages show what they make. Each preset carries a long description, use cases and example media you can run. - **Billing.** Top up credits with one control where you choose the exact amount. - **Editor.** Media clips in the editor have Motion controls for position, scale, rotation, opacity and anchor, with on screen handles in the monitor. - **Web.** Every preset has a shareable public page with a launch button, plus a short /p link for campaigns. - **API, Editor.** Per-clip motion in Studio. Position, scale, rotation, opacity and anchor on any clip, keyframed over time, rendered identically in the editor and the exported MP4. - **Agent.** Switching the NOLGIA Agent brain shows live progress in the chat, and your message sends when the switch finishes. - **Editor.** Studio columns and the monitor to timeline split are resizable, and your layout is remembered. - **API, Agent.** The agent read reports brain switch progress, and messages sent during a switch are queued instead of failing. - **Editor.** The Studio monitor puts preview quality in the transport bar, keeps its controls in fullscreen, and shows live shortcut tooltips. - **Web, Agent.** You can rate NOLGIA Agent replies with a thumbs up or down and share finished images and videos with a durable link. - **Editor.** Project Media and Library offer a list view with sorting, you can create a new timeline inside the editor, the toolbar never scrolls, and adding an audio track is one click. - **Agent.** The NOLGIA Agent shows its name and current brain in the chat header, and its activity reads in plain language. - **Web.** Generation progress shows on every page, thumbnails stay sharp after a backgrounded tab, and a Notify me toggle tells you when a run is ready. - **Web, Agent.** Start creating and the landing page prompt box open a fresh NOLGIA Agent chat right away. - **Agent.** Every new Agent starts with the core ability set installed; agents created earlier without abilities were given the same set. - **API.** Share links for assets and renders are durable and revocable through the API. - **Web, Agent.** Rate an Agent reply. Give any reply a thumbs up or down; your feedback is stored with the reply and reviewed. - **Editor.** Project settings let you change the frame rate, and editor shortcuts are customizable with Premiere Pro and CapCut presets. - **Web.** You can sign in with Apple, and your session stays signed in across tabs and reloads. - **Web.** A left rail follows you through the app with Create video, Timeline editor, Presets, Library and Projects, plus a Continue where you left off section that remembers your last chat and timeline. - **Agent.** Every NOLGIA Agent runs the NOLGIA persona with a customer facing conversation style. ## Week of August 31, 2026 **September 4, 2026** - **API.** Image jobs on queued routes get up to 10 minutes to finish, and a content moderation rejection shows a clear message instead of a generic failure. - **API.** Kling jobs that cannot start fail within seconds with a refund instead of waiting 15 minutes for a timeout. - **Agent.** The Agent watches video. Attach or link a clip and the Agent can look at it and reason about what it sees, on Gemini 3.8 Flash. - **API, Agent.** Gemini 3.8 Flash can be selected as a NOLGIA Agent brain through the API at 25 credits per turn. - **Agent, Billing.** Gemini 3.8 Flash is available as a NOLGIA Agent brain at 25 credits per turn. - **Web.** Reve 2.1, Seedream 5.0 Pro, Ideogram v4 and Qwen Image 3 are available for image generation and editing. - **API.** Reve 2.1, Seedream 5 Pro, Ideogram v4 and Qwen Image 3 are available through the API for image generation and editing. - **Web.** Nine new models joined the catalog: Nano Banana Pro and 2 Lite, GPT Image 1.5, Grok Imagine 2.0, Veo 3.1 Lite, Eleven v3 Conversational, HappyHorse 1.1, and Seedance 2.0 Fast and Mini. - **API.** Nine new models are available through the API, including Nano Banana Pro, GPT Image 1.5, Grok Imagine 2.0, Veo 3.1 Lite and Eleven v3 Conversational. - **API.** More image models. Nano Banana Pro, Nano Banana 2 Lite, GPT Image 1.5 and Grok Imagine Image join the catalog on direct provider routes, alongside Reve 2.1, Seedream 5 Pro, Ideogram v4 and Qwen Image 3. **September 3, 2026** - **Web.** Model pickers now list models best first by the current leaderboards, with Seedance 2.5 the default for video and ElevenLabs v3 for voiceover. - **API.** Reusable reference elements. Save a reference image once and attach it to any generation with element_ids; models that take element images receive them in the right slot. - **Web, Editor.** Restore footage from Studio. De-noise, de-haze and up-res an existing clip with restore engines chosen by the problem in the footage. - **Web.** Seedance 2.5 is available for video generation, covering text to video, first and last frame animation, and reference stills. - **API.** Seedance 2.5 is available through the API for text to video, first and last frame, and reference based generation. - **Web.** Google Drive and your own storage. Import media from Google Drive into your Library and export finished work back (Studio plans and above). Enterprise organizations can connect S3-compatible storage and mirror every asset there. - **Web, Billing.** Team plans are seat-based. Choose seats at checkout, add or remove seats later with prorated billing, and pool credits across the whole team. - **Web, API.** Teams and Enterprise organizations. Invite members with roles, share one Library and one credit pool, verify your domain so colleagues join automatically, and create organization API keys. Switch between your personal space and an organization from the account menu. - **Web.** Enterprise SSO. Organizations can sign in with SAML or OIDC, enforce it for every verified domain, and auto-join members from that domain. ## Read it programmatically The same entries are published as JSON. Each one carries the day it shipped, the copy above and the surfaces it touched, and nothing else. ```bash tab="curl" $ curl -sS https://nolgia.ai/changelog/entries.json ``` This page is generated from that document when the docs publish, so the two cannot disagree.