CreatorOS Studio rebuilds open-source Higgsfield on ComfyUI: photo restore, in-UI character LoRA training, 60+ tools, local and free

Key_Cap_8111 · reddit · 2026-11-04

For the extended ComfyUI 'Open Source Higgsfield' challenge, the author shipped a major CreatorOS Studio update — a web app wrapping a local ComfyUI.

• Restore: old photos repaired + colorized (Qwen-Image-Edit 2511) then rebuilt in HD (SeedVR2); damaged clips to 1080p with SeedVR2 7B; experimental full-clip colorization via Wan 2.2 Fun VACE from one frame (the 1951 newsreel demo).
• Train characters in the UI: describe, pick from 4 faces, get a reference sheet, 12 images and a LoRA on FLUX.2 Klein / Krea 2 (ai-toolkit) with live progress and cancel; each character gets a voice.
• One brief → one reel: shots 2..n generated from shot 1's photo for a single-session look; 36 style mosaics, 10 relights, blurred-prefill expansion; one-page one-step tools (inpaint, remove background, seamless loop, HDR masters...); timeline editor, CLI + MCP; every render downloads its exact workflow.

Limits: built and measured on a 5090; 24GB VRAM for nearly everything; public demo runs one SFW render at a time. Results Nov 10.

Original post →

More from Multimodal

• Hedra speeds up Krea 3 diffusion transformer 6.9x on AMD Instinct MI355X — henloitsjoyce · 2026-11-04
• Pruna open-sources P-H3-Image, converting SOTA video model MiniMax-H3 into an image model — multimodalart · 2026-11-04
• PrunaAI Open-Sources P-H3-Image: 7x Smaller, 5x Faster, 8-Step Images — sdiazlor · 2026-11-04
• PrunaAI Open-Sources P-H3-Image: 7x Smaller, 5x Faster, 8-Step Images — sdiazlor · 2026-11-04
• AssemblyAI launches first TTS model on LiveKit: phone-call-trained, 90ms first audio, $15/1M chars — AssemblyAI · 2026-11-04
• From one 82-second video to a 3D Gaussian splat in ~20 minutes — janusch_patas · 2026-11-04

Multimodal channel →