ReDeck: Framework for Document-to-Slide Generation Refinement
A framework for step-level render-grounded refinement in document-to-slide generation.
ReDeck is a framework that decomposes slide revision into atomic edit actions and returns renderer-derived observations after each step. It uses multi-granular feedback: step-level render feedback for spatial errors, turn-level adaptive critic for semantic and design guidance, and submission-level gate for hard layout validation.
Improving the accuracy and quality of AI-generated slides from documents, addressing issues like overflow, overlap, clipping, and off-canvas placement.
Not specified
http://arxiv.org/abs/2609.00194v1
ReDeck is a framework for refining document-to-slide generation. It breaks down slide revisions into small steps, providing feedback after each edit. This helps fix spatial and layout errors more effectively than traditional methods.
By providing feedback at each edit step (one edit, one observation), ReDeck allows for immediate identification and correction of local failures like overflow or clipping, which are hard to fix with delayed feedback.
When the document is very short or simple, and direct generation without iterative refinement is sufficient.
- ReDeck: Step-Level Render-Grounded Refinement for Document-to-Slide Generation ↗arXiv · arXiv cs.AI · Aug 31, 2026 · paraphrased
First seen on Subvolts Aug 31, 2026. Updated Sep 11, 2026. Spotted a problem? Tell us.
- ReDeck: Step-Level Render-Grounded Refinement for Document-to-Slide GenerationAnthropic · arXiv cs.AI · Aug 31, 2026
- ReDeck: Step-Level Render-Grounded Refinement for Document-to-Slide GenerationGoogle · arXiv cs.AI · Aug 31, 2026
- ReDeck: Step-Level Render-Grounded Refinement for Document-to-Slide GenerationOpenAI · arXiv cs.AI · Aug 31, 2026
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Troubleshooting fixDecrease LM discrimination via careful prompt engineeringUse prompt engineering to significantly decrease positive and negative discrimination in LM outputs. · 20 min
- ChecklistAudit LLM evaluation for citation artifacts and biasCheck for presentation cues, citation issues, and quality gaps in LLM-as-a-judge evaluations. · 30 min
- Eval methodCheck LLM state policy recall for cross-jurisdiction substitutionVerify if an LLM provides another state's policy value when asked about a specific state. · 60 min
- ChecklistLimitations of text watermarking detectionCheck for factors that can affect text watermark detection reliability. · 2 min
- TemplateEcological feature inventory and planning briefTemplate for documenting and planning the use of fine-scale ecological feature data. · 20 min
- HeuristicStep-level feedback for slide generation refinementProvide feedback after each small edit, not just at the end of a full revision cycle. · 10 min
Questions people ask
- Which AI tools does this work with?
- Any current model. It was seen with in the sources.
- Where did this come from?
- arXiv (arXiv cs.AI). Subvolts collected, normalized and credited it; the link above opens the original.