The four online tools now open with a real example already in the try-it panel: the source image is the default reference, and the result is the default preview.
Changelog
Grouped by version, newest first. Only changes you can actually use on the site.
v1.5
Online tools, docs, and Grok
The Claude Code guide was rewritten around a fresh computer: native install, environment variables, how to confirm the model list, and what counts as a successful first request.
The English docs now include getting started, authentication, API keys, the wallet and task lookup. The English home page no longer says the docs are still being written.
Grok Build can now run on ModelFold chat models, including tool calls and web search. The docs cover setup and context length.
Conflicting inputs on some video models are now rejected before generation. You cannot send both a document and a web link, or mix start and end frames with reference video, so the job fails up front instead of after you submit.
Online tool pages now use the same try-it layout as model pages, and each one can show your past results for that tool. All four tools are in the top navigation and the footer.
Four online tools are live: remove or replace a background, upscale an image, make a headshot or avatar, and turn a photo into a video. You can read and try the form immediately; generation spends credits for the model it uses.
The language menu now says Simplified Chinese. The pricing page compares our credits with publicly listed provider prices.
v1.4
Sign-in, market, languages, and playground
The site is now available in Spanish. Public pages have an /es/ address, and a Spanish browser opens the Spanish site. Long model articles still fall back to English for now.
Model playgrounds have a Reset button. It clears the form and uploaded files, and keeps the job in progress plus any result you already have.
The site is now available in Japanese. The interface, model summaries and parameter labels are translated. Long articles are in Japanese for the popular models; the rest still show English.
The model market now pages through results, autoplays its covers, and uses sharper artwork. Newly added models appear near the front.
If the browser is already signed in to Google, the site can offer one-tap sign-in instead of sending you through the login form.
v1.3
Alipay, apps, and new models
Kling 4.0 is available for reference-to-video and keyframe-to-video. The model page and the API docs are both up.
Seedance 2.0 Mini is available, alongside Seedance 2.0 and 2.5. Each has a model page with pricing and a playground.
A batch of model pages now shows examples, prompts and previews that match the model itself. Related recommendations no longer include models that are not actually available.
AI Interviewer and AI Model Advisor are live. Interviews can be saved and downloaded, and recommendations keep a history. You can open both from the home page, the market, pricing and the API key page.
The wallet now accepts Alipay, in addition to cards, Apple Pay, Google Pay and WeChat Pay.
v1.2
AI Meeting, invites, and background jobs
Flux 2 image-to-image now accepts up to 8 reference images in one request.
Image, video and audio playground jobs now run in the background. You can leave the page and download the result later from your usage history. Starting prices in the model list use the real lowest tier.
Wallet transactions can be filtered by type and date. Usage can be filtered by model, status and date. Filters and page stay in the address, so a refresh keeps your place.
You can invite friends. Share your link or code; when they sign up, you both receive bonus credits on top of the regular sign-up bonus. The link opens the sign-up page with the code filled in.
AI Meeting, a desktop app, is available for Mac. It turns audio playing on the computer into live captions, spots questions, and drafts answers with your ModelFold chat model. There are Apple silicon and Intel builds. The app is free; answers spend credits.
v1.1
Developer access and the Chinese site
Model names on the home page and the pricing page link to the matching model page, and the credits shown match what is actually charged. You can top up with a pack or type your own amount.
A cookie choice is now shown. Sign-in storage stays on; language and theme can be accepted separately. You can reopen the choice from the footer.
Public pages now have Chinese addresses, such as /zh/ for the home page. A first visit follows the browser language; after you switch, the site and the docs remember it.
New accounts receive sign-up credits once, whether you register with email, Google or GitHub. The amount is whatever lands in the wallet at sign-up.
Claude Code and Codex CLI can call ModelFold chat models through their own native APIs, with streaming and tool calls. The docs have a separate AI Agent section.
v1.0
Launch: models, payments, and billing
Chat, image and video models now have detail pages with an overview, pricing and a playground. Video includes Seedance 2.5 and Seedance 2.0. Open them from the model market.
Video and image calls are priced by tier, such as resolution and duration, and the playground shows the credits before you submit. Failed generations are refunded. You can also type a custom top-up amount.
The wallet accepts cards, Apple Pay, Google Pay and WeChat Pay. Credits are added to the account after payment.
ModelFold is available at ai.modelfold.online. You can sign up with an email code, Google or GitHub. The terms and the privacy policy are published.