Research map / Image generation / Text-to-image diffusion
Dual-Process Image Generation
Grace Luo, Jonathan Granskog, Aleksander Holynski, Trevor Darrell
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 |
|---|---|---|---|
| apply dual-process VLM distillation on top of the base prompt | Flux Schnell · CommonsenseT2I | accuracy (%): 24.4 -> 44.5 | Improved |
| apply the method on top of GPT4o prompt expansion | Flux Schnell · CommonsenseT2I | accuracy (%): 50.7 -> 62.0 | Improved |
| apply the method on top of expanded prompt | Flux Dev · CommonsenseT2I | accuracy (%): 44.2 -> 58.2 | Improved |
| apply the method on top of expanded prompt | Flux Schnell · CommonsenseT2I Animal Behaviors | accuracy (%): 70.8 -> 66.7 | Hurt |
| use more reliable questions derived from ground-truth inferences (ablation) | Ours (Flux Schnell) · CommonsenseT2I | accuracy delta from No Control (%): +20.1 -> +28.2 | Improved |
| supervise with Idefics2 instead of LLaVA (ablation) | Ours · CommonsenseT2I | accuracy (%): 14.5% boost | Improved |
| apply the method to Flux Schnell generations | Flux Schnell · CommonsenseT2I image quality | FID: 102.8 -> 98.7 | Improved |
| optimize weights instead of latents (ablation) | single-seed VLM loss optimization · Computational cost | inference time: 56s -> 0.3s | Improved |
| optimize weights instead of latents (ablation) | single-seed VLM loss optimization · Computational cost | Image generator VRAM: 42GB -> 44GB | Hurt |
| use the method for horizon control versus prompt expansion | Flux Schnell · Horizon control | horizon distance: 0.1923 -> 0.1705 | Improved |
Closest papers by meaning
- VIGC: Visual Instruction Generation and Correction (AAAI 2024 · similarity 0.86)
- DuoGen: Towards Autonomous Interleaved Multimodal Generation (CVPR 2026 · similarity 0.86)
- Instruct-Imagen: Image Generation with Multi-modal Instruction (CVPR 2024 · similarity 0.85)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.85)
- Synthesizing Visual Concepts as Vision-Language Programs (CVPR 2026 · similarity 0.85)
- Kosmos-G: Generating Images in Context with Multimodal Large Language Models (ICLR 2024 · similarity 0.85)
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.
