AI

Controllable Diverse Sampling through Pullback Directions

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 14/10/2026, 12:19:27 UTC.

Briefings

A new training-free method for diverse image generation uses local pullback geometry to identify sampling directions that maximize output sensitivity. Experiments on MS-COCO and BrushNet demonstrate improved diversity and consistency in prompt-conditioned image generation and inpainting tasks.

Key points

• The method leverages local pullback geometry induced by the mapping from conditioning-embedding space to image space to identify directions of maximal output sensitivity.
• Steering sampling trajectories along these directions encourages distinct generations while maintaining consistency with the conditioning signal.

Context

Conditional diffusion models generate high-quality images aligned with prompts but often produce visually similar samples from different seeds. This method offers a training-free approach to enhance diversity without additional optimization overhead.

Limitations

Results are limited to MS-COCO and BrushNet datasets. No additional background context is supplied beyond the abstract; preprint status and uncertainty remain.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 25/10/2026, 19:16:30 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiaigcm7p2c2qcudqkvj2saayplozdqqkmulpcgnodbqif74cx2gs4