Higgsfield's 120 creative apps, rebuilt on open weights and ComfyUI
Freefield clones the AI-effects studio's app lineup as ComfyUI workflows you can run on Modal or your own GPU — no API nodes, no per-app code.

What it does Freefield rebuilds Higgsfield’s image, video, VFX and audio apps — 120 of them — as ComfyUI workflows behind a web UI laid out like the original. Every model is open weights downloaded from Hugging Face, with size and license listed per app. It runs on Modal or on your own ComfyUI, and the heavier video apps (LTX-2.5, MiniMax H3) mostly need cloud GPUs anyway.
The interesting bit
There is no per-app code. Each app is just a folder with an app.json manifest and a workflow in ComfyUI’s API format; the UI forms, runner, model downloads, VRAM map and tests are all generated from those manifests. tools/import_workflow.py drafts a new app from any ComfyUI workflow, so the whole studio is data, not code.
Key highlights
- 119 ready apps: 39 with their own workflow, 72 template apps (a fixed prompt over a base pipeline), plus smoke apps; every model-using app ships a real GPU example with its runtime and peak VRAM.
- 48 lab runs on Modal (L40S/H100), 710 of 733 cases ok; end-to-end timings recorded, e.g.
morphat 160.8 s button-to-video. - A 334-case LoRA lab with fixed seeds — one showcase per app plus experiments, each with a written verdict.
- CI validates every manifest, workflow and lab prompt against a CPU-only ComfyUI in about 3 minutes.
- Honest VRAM accounting: each app states whether it fits a 24 GB card, fits by phases, or needs Modal.
Caveats
- Many of the best apps (cinema video, dub, reshoot, H3 video) need Modal-class GPUs — a 24 GB card only gets you partway.
moss-imageis still a draft, and 72 of the apps are prompt templates over shared pipelines rather than distinct workflows.- The lab found rough edges it documents openly, e.g. the order of two camera moves in
cinema-videois not reliable.
Verdict Worth a look if you want Higgsfield-style creative tools without the subscription or closed models, and you don’t mind renting GPUs for the video half. Skip it if you only have a 24 GB card and wanted the cinematic video apps — those live in the cloud.
Frequently asked
- What is neuregex/freefield?
- Freefield clones the AI-effects studio's app lineup as ComfyUI workflows you can run on Modal or your own GPU — no API nodes, no per-app code.
- Is freefield open source?
- Yes — neuregex/freefield is open source, released under the MIT license.
- What language is freefield written in?
- neuregex/freefield is primarily written in Python.
- How popular is freefield?
- neuregex/freefield has 1.1k stars on GitHub.
- Where can I find freefield?
- neuregex/freefield is on GitHub at https://github.com/neuregex/freefield.