Paste or pick your prompts
Bulk import a script, with a line of --- between scenes, and each scene becomes its own row. Or add rows one by one, or pick from your saved prompt lists.
Paste a whole script of prompts, or pick them from your saved lists, and ZumiForge makes each one its own image or video. Give any scene its own reference image, press Queue once, and come back to a finished set. Free models run on your Mac; cloud models run with your own keys.






Six prompts, one press of Queue. Seedream 5 Pro through fal.ai.
Storyboards, picture books, product shots, social posts, a set of thumbnails: anything that's many pictures rather than one. ZumiForge keeps the whole set together, so you write it once and let it run.
Bulk import a script, with a line of --- between scenes, and each scene becomes its own row. Or add rows one by one, or pick from your saved prompt lists.
Add reference images for every prompt, or drop one on a single scene so only that one uses it. For video, a scene's own image becomes its first frame. Templates keep a character or place described the same way in every scene.
Every scene goes into the queue. Your Mac and each cloud service work through it at the same time, the results fill the gallery as they finish, and anything that fails has a Retry button that sends the exact same job again.






Another six-scene story, queued in one go: MiniMax H3 Image, directly with a MiniMax key.
ZumiForge is a Mac app with its own window and a small helper in your menu bar. The helper runs the models on your Mac and talks to the cloud services for you.
Every prompt has its own row. Add, remove and edit them, paste a long script to split it into rows, and the list is still there after you queue, ready to tweak and run again.
Save prompts in named lists, import a script of scenes, search them, and send any selection back to the page. Prompts you've queued are marked, so you know what's done.
Describe a character, place, outfit or lighting once and use it in any prompt with a tag like {{character:marcus}}. Add reference images for the whole set or for single scenes.
Your Mac and each cloud service work at the same time, so a long cloud video never holds up a quick local image. Stop or remove any job, and retry a failed one exactly as it was.
Switch to Video and the same list of prompts becomes a set of clips, with length, quality and sound. A scene's reference image becomes its first frame.
Z-Image Turbo, Flux2 Klein, FLUX.3 Klein, Qwen-Image, Krea, Iris, Ideogram, MiniMax H3 Image (with multi-image editing) and LTX-2.5 video run on Apple silicon. Nothing leaves your computer.
FLUX, Seedream, GPT Image, Nano Banana, Grok Imagine, MiniMax H3, Kling, Veo, Seedance and more, through fal.ai, Replicate, Runware and Higgsfield, or directly from Google, OpenAI, xAI and MiniMax.
The summary bar says how many images or videos you'll get and estimates the total from list prices where services publish them. Each finished card shows what it cost.
Search everything you've made by prompt, model, type or status, page through it, and copy any prompt back to try again.
Every prompt goes to each model you pick, so the same scene comes back in each one's style and you can see which fits before you run the whole set. Here is scene 1 of the Paris story, "The wait", from four of them.




Needs a Mac with Apple silicon (M1 or newer) and macOS 13 or later. Open Terminal and paste:
curl -fsSL https://zumilabs.com/tools/zumiforge/install.sh | sh~/Pictures/ZumiForge; change it any time from ☰ → Change image folder, and open it with ☰ → Show in Finder.Prefer to download it yourself? ZumiForge-mac.zip (about 55 MB, SHA-256). ZumiForge isn't signed with an Apple Developer ID yet, so a copy downloaded in a browser needs System Settings → Privacy & Security → Open Anyway the first time. The Terminal install above avoids that warning.
As many as you like. Your Mac runs one local job at a time (they share its memory), while each cloud service runs three at a time by default, all side by side. With several models ticked, each prompt is made once per model, and the Queue button says the total before you press it.
ZumiForge is free. Models on your Mac are free to run. Cloud models are billed by each service on your own account; ZumiForge shows list prices and what each job cost.
Images and videos go to the folder you choose. Your history, prompt lists and templates stay in ~/Library/Application Support/ZumiForge. API keys are kept in the macOS Keychain. When you use a cloud model, that service receives the prompt and any reference images for that job. ZumiLabs receives nothing. See the privacy policy.
Each is a one-time download of about 6–70 GB from Hugging Face, and some need you to accept the model's licence there first. The built-in walkthrough checks your Mac, helps with the Hugging Face token, and downloads and tests each model. Cloud-only use needs no downloads.
Apple silicon Macs (M1 or newer) with macOS 13 or later. 32 GB of memory is comfortable for local models, and 16 GB works for the smaller ones. Cloud models work on any of them.
ZumiForge isn't signed with an Apple Developer ID yet. Installed through Terminal, macOS doesn't mark it as downloaded from the internet, so it opens without warnings. The installer checks the download against its published checksum.
Run the install command again to update; your history, settings and keys are kept. To remove it, quit ZumiForge from its menu bar icon and move ZumiForge from Applications to the Bin. Your images stay in their folder; delete ~/Library/Application Support/ZumiForge to remove your history too.