Research map / Image generation / Autoregressive and latent generation
The Latent Color Subspace: Emergent Order in High-Dimensional Chaos
Mateusz Pach, Jessica Bader, Quentin Bouniot, Serge Belongie, Zeynep Akata
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Inject color via LCS global intervention instead of no color in prompt | FLUX.1 [Dev] · GenEval color task | accuracy: 9% -> 72% | Improved |
| Inject color via LCS global intervention instead of prompting | FLUX.1 [Dev] · GenEval color task | accuracy: 79% -> 72% | Hurt |
| Restrict LCS intervention to target object (local) instead of global (ablation) | LCS · GenEval color task | accuracy: 72% -> 68% | Hurt |
| Use LCS intervention instead of prompt | FLUX.1 [Dev] · Precise (plain) | CIEDE2000: 20 -> 11 | Improved |
| Use LCS intervention instead of prompt | FLUX.1 [Dev] · Precise (plain) | hue error: 24 -> 8 degrees | Improved |
| Use local LCS intervention instead of prompt | FLUX.1 [Dev] · Precise (natural) | CIEDE2000: 23 -> 14 | Improved |
| Use LCS instead of prompting for structure preservation | FLUX.1 [Dev] · GenEval color task | SSIM: 0.46 -> 0.59 (local), 0.46 -> 0.56 (global) | Improved |
| Use local LCS instead of best-of-50 sampling | FLUX.1 [Dev] · Precise (natural, small) | CIEDE2000: 19 -> 10 | Improved |
| Use local LCS instead of ReNO, ColorPeel and IP-Adapter | FLUX.1 [Dev] · Precise (natural, small) | CIEDE2000: 23 / 29 / 40 -> 10 | Improved |
| Use interpolation (Type star) instead of direct LCS shift (Type I) (ablation) | LCS intervention · FLUX.1 [Dev] qualitative | texture preservation | Improved |
| Use interpolation (Type star) instead of HSL shift (Type II) at early timesteps (ablation) | LCS intervention · FLUX.1 [Dev] qualitative | color influence | Improved |
| Apply LCS intervention later (t=20-25) with cleaner boxes (ablation) | LCS intervention · FLUX.1 [Dev] qualitative | fine-grained detail preservation | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ColorPeel — ColorPeel: Color Prompt Learning with Diffusion Models via Color and Shape Disentanglement (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective (NeurIPS 2024 · similarity 0.87)
- Color Alignment in Diffusion (CVPR 2025 · similarity 0.86)
- Improving the Diffusability of Autoencoders (ICML 2025 · similarity 0.86)
- Angle Domain Guidance: Latent Diffusion Requires Rotation Rather Than Extrapolation (ICML 2025 · similarity 0.84)
- Masked Autoencoders Are Effective Tokenizers for Diffusion Models (ICML 2025 · similarity 0.84)
- Free-Lunch Color-Texture Disentanglement for Stylized Image Generation (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
