New models, improvements, and fixes. Every update, documented.
No updates yet.
Make 2 to 30 second videos at 480p, 720p or 1080p, with native sound: dialogue, music and effects. Reference mode takes up to 10 images to guide the shot, and First & Last animates from a first frame to an optional last one. Prime renders faster for about 40% more.
See all models →Wan 3.0 and Wan 3.0 Prime gain Edit and Extend. Edit changes a clip and keeps its length and shape. Extend continues one to the total length you set. References can now be up to 10 images, 5 clips, 5 sounds, and one document or public web page.
Set a daily credit limit for each app or agent connected to your account, or pause one. Agents that can't open a sign-in page can show you a short code to enter at app.fluxpool.ai/verify.
Seedance 2.0 Fast and Seedance 2.0 Mini give you Seedance 2.0 for less, at 480p or 720p, with the same modes, references and sound. Flux 2 Klein is the fastest and lowest-priced image model on Fluxpool. Flux 2 Max is the top Flux model, and when a prompt asks about something current it looks it up on the web before drawing.
See per-model pricing →Turn on Draft to render a quick 480p version for less. When you like one, render the final at 1080p with the same prompt and settings. Drafts are kept for 7 days.
Seedream 5.0 Pro and Flash have a Layers mode that splits one image into separate layers. View them, download them one at a time, or export them all as a ZIP.
On Flux 2 Pro, attach an image, brush over the area to regenerate and describe what goes there. The rest of the image stays as it was.
Pick what you want Seedance to do. Reference makes a video from a prompt, optionally guided by references. First & Last animates between a start frame and an optional end frame. On Seedance 2.5, Extend continues a clip you already have and Edit changes something inside it. Sound can be turned off for a silent video at the same price.
Seedream 5.0 at a fraction of Pro's price and faster, with slightly less detail. Same sizes and up to 10 reference images.
In Claude, add Fluxpool as a custom connector with the address https://mcp.fluxpool.ai/mcp, then approve it in Fluxpool. There is no key to copy, and the approval screen shows which account's credits it will use. API keys keep working for Cursor, Claude Code and other MCP clients.
On the Hero plan, connect a Google Drive folder and finished generations are saved there as well. Pause syncing or disconnect at any time.
Give it 1 to 5 reference images and it keeps the subject's identity while your prompt drives the motion and camera. 720p or 1080p, up to 10 seconds.
Credits from plans and packs are now one balance. New accounts start on a 1-month Trial plan with 50 credits and every model unlocked. No credit card needed.
Connect Claude Desktop, Cursor, or any Model Context Protocol client to Fluxpool through our hosted MCP server, now at https://mcp.fluxpool.ai/mcp. Six tools ship out of the box: image generation, video generation, credit balance, library browse, recent prompts, and model catalog.
Select multiple assets in Library, then download or delete them together from the new Actions dropdown. Collections now open directly to a filtered Library view.
The Generate page now keeps your prompt and parameters after a generation completes so you can iterate quickly. The seed used is shown on the result, and the "Use as input" panel is expanded by default.
Two new sections on the Generate page for image models: pick a resolution tier plus aspect ratio to see the exact pixel dimensions, and choose PNG or JPEG per generation.
If the balance can't be fetched, the dashboard now surfaces the error instead of silently showing 0. Library downloads use a fetch-plus-blob path for reliable cross-origin saves.
Generated assets are retained for 14 days. Library now shows an "Expires in X days" row on each asset, badges assets nearing expiry, and hides fully expired ones — so nothing silently disappears on you.
Announcements posted from the admin panel now appear as a banner above the dashboard nav, so users see incidents, new-model launches, and other timely updates on their first screen.
Fixed image upload on the Generate page including drag-and-drop, and your prompt is now preserved when you switch between models.
Error toasts no longer leak raw internal messages — you get a sentence you can act on. Added skeleton loading states across the dashboard, credit balance, and usage charts so the app feels instant on cold loads.
The Generate button now shows how many parallel jobs your plan allows and how many you have running, so you can pace bursts.
A prompt-history dropdown on the Generate page recalls recent prompts — cycle with ↑ / ↓ arrow keys. In-flight generations now reconnect after a page refresh instead of appearing to vanish.
Free users are treated like paid users: same per-plan concurrency limits, same account settings shape. Removes the second-class behaviour where certain features weren't reachable on the free tier.
Retry any failed generation directly from Library. Use any output as the input for the next one. New example-prompts panel makes it easier to explore what a model can do.
Library now shows generations that are still running or that failed, alongside the finished ones — you always know the state of every job you kicked off.
ByteDance's next-generation video model. Up to 30-second single-pass generations, native audio, and expanded reference media (30 images / 10 videos / 10 audio). Model ID: seedance-2.5. Available on the playground and via the API.
Video pricing (Seedance, Wan, Kling) is now computed live from your parameters — resolution, duration, and whether you're passing an input video — instead of a fixed per-tier price. The Generate button shows the exact cost as you change inputs.
Dola-Seedream 5.0 Pro now supports image-to-image with up to 10 reference images. Model ID: seedream-5.0.
Invite teammates into a shared organization: one credit pool, org-shared Library with creator chips on each asset, role-based delete, per-org Stripe billing, and an audit log page. Switch between personal and team contexts from the dashboard.
One click on any asset in Library reopens the Generate page pre-filled with the same prompt, model, and parameters — ready for you to tweak and regenerate.
API responses now return structured error codes you can branch on programmatically. The Generate page shows the exact cost of the next generation before you click, based on your current parameters.
All API endpoints now serve from api.fluxpool.ai. Existing callers pointing at the older *.zkcandy.io hosts will need to update their base URL. See the API reference for the current endpoints.
Ported the LLM chat surface into the dashboard alongside image and video. (LLM models are wired progressively as backend endpoints come online.)
Sidebar toggle for smaller screens, deduplicated top nav, and a collapsible model selector so you can hide the picker to see more output.
Added ByteDance Seedream 4.0 to the image model catalog. (Superseded by Seedream 5.0 in early August.)
The dashboard now respects your browser language and ships with English and Japanese translations. Voucher redemption UI added.
Generate accepts up to 9 reference images for Seedance. Polling now waits up to ~25 minutes for long video jobs so results don't get lost.
Raised the prompt-box limit from ~2,000 characters to 5,000 words. Generate now supports running multiple generations at the same time — a Recent Generations grid tracks them all.
We use cookies that are strictly necessary to run Fluxpool, and — only with your permission — analytics and marketing cookies that tell us which channels bring people here. See our Privacy Policy.