Latent Space Color Editing Breakthrough in Krea2: ComfyUI Node Coming
muerrilla · reddit · 2026-07-30
Developer muerrilla accidentally discovered powerful color adjustments using extracted latent space vectors while experimenting with Krea2 (Qwen Image VAE). Using pure vector math—without LoRAs, complex prompts, or CFG boosts—this method achieves photographic color editing akin to Camera Raw, handling exposure, temperature, tint, and contrast.
Because this operates directly in the latent space during sampling, it offers a very high dynamic range and can steer the diffusion process beyond the model's normal limits (e.g., generating extremely dark or bright images, or influencing morphology).
The developer is turning this into a user-friendly ComfyUI custom node featuring sliders and range masking tools. Vectors for Flux VAE (ZImage) are also being extracted, meaning the node could potentially support other models sharing the same VAE architecture.
More from Multimodal
- Seedance 2.5 workflow: character reference sheets + structured prompt yields a 30s single-take fake tourist video — techhalla · 2026-10-16
- AI couple moves into a castle ruin: full Magnific + Seedance 2.5 prompt thread behind the viral clip — techhalla · 2026-10-16
- FLUX.3 Klein shines at consistent high-res photo editing on RTX 5090: 3000×4000 in ~230s — herbertseabra · 2026-10-16
- Smoky Velvet Noir: a reusable image-gen prompt template with three moody examples — LudovicCreator · 2026-10-16
- VideoGen-Agent: An 8B Multimodal Agent Turns Video Generation Into a Tool-Use Loop — jiqizhixin · 2026-10-16
- Redditor builds impressive seaplane sim project with Sonnet 5.5 + Blender — Wise_Shame_2853 · 2026-10-16