Research map / Image generation / Text-to-image diffusion
Interpretable Prompts made Edit-Friendly: Token-to-Token Similarity Reduction in dLLMs for Edit-Friendly Hard Prompt Inversion
Naresh Kumar Devulapally, Shruti Agarwal, Vishal Asnani, Vishnu Suresh Lokhande
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 |
|---|---|---|---|
| add CLIP steering to plain dLLM decoder (ablation) | dLLM prompt inversion · MS COCO (32 tokens) | CLIP image similarity: 0.62 -> 0.68 | Improved |
| add CLIP steering to plain dLLM decoder (ablation) | dLLM prompt inversion · MS COCO (32 tokens) | perplexity: 36.91 -> 28.84 | Improved |
| add token-token similarity reduction on top of CLIP guidance (ablation) | dLLM + CLIP Guide · MS COCO (32 tokens) | CLIP image similarity: 0.68 -> 0.71 | Improved |
| add token-token similarity reduction on top of CLIP guidance (ablation) | dLLM + CLIP Guide · MS COCO (32 tokens) | perplexity: 28.84 -> 33.16 | Hurt |
| add token-token similarity reduction on top of CLIP guidance (ablation) | dLLM + CLIP Guide · MS COCO (32 tokens) | TIFA: 0.81 -> 0.89 | Improved |
| add token-token similarity reduction on top of CLIP guidance (ablation) | dLLM + CLIP Guide · MS COCO (32 tokens) | inference time (s): 3.42 -> 5.60 | Hurt |
| token similarity reduction without CLIP guidance (ablation) | dLLM prompt inversion · MS COCO (32 tokens) | TIFA: TIFA 0.74 -> 0.92 but PPL 36.91 -> 45.93 | Mixed |
| replace autoregressive beam search with CLIP-guided dLLM decoding | hard prompt inversion · Inversion time (77 tokens) | time (s): 104.20 -> 10.50 (VGD vs Ours) | Improved |
| use dLLM + CLIP + similarity steering instead of VGD | hard prompt inversion · MS COCO (16 tokens) | perplexity: 100.17 -> 54.68 | Improved |
| use our method instead of VGD on JourneyDB (CLIP text and F1) | hard prompt inversion · JourneyDB (32 tokens) | CLIP text similarity: 0.48 -> 0.48 | No clear effect |
| increase token budget for inversion | Ours · MS COCO | perplexity: 54.68 -> 17.10 | Improved |
Closest papers by meaning
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.91)
- Visually Guided Decoding: Gradient-Free Hard Prompt Inversion with Language Models (ICLR 2025 · similarity 0.90)
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.89)
- Who Evaluates the Evaluations? Objectively Scoring Text-to-Image Prompt Coherence Metrics with T2IScoreScore (TS2) (NeurIPS 2024 · similarity 0.88)
- EditInfinity: Image Editing with Binary-Quantized Generative Models (NeurIPS 2025 · similarity 0.87)
- DynT2I-Eval: A Dynamic Evaluation Framework for Text-to-Image Models (NeurIPS 2026 · similarity 0.87)
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.
