longlian/llm-grounded-diffusion

Update requirements.txt

Update requirements.txt

Fix checkbox label

Pin the gradio client version

Add pydantic versioning to requirements.txt

Use fast_after_steps only with use_fast_schedule

Update placeholders and fix defaults with low_memory

Update instructions

Update the labels to examples

Fix default preset with low_memory

Update implementation note

Update instructions

Improve error handling in parsing

Update cached examples

Update: add attention guidance and refactor the code

Close the plot in SAM

Use fast schedule for per-box generation to speed up

Update examples

Update examples

Apply batching to SAM to reduce the memory cost with many objects

Empty cache after each generation

Empty GPU cache at the end of run

Reduce so_batch_size for lower memory cost

Update default steps

Update examples

Add batched single object generation

Fix scaling in input latents_bg

Add autocast

Update the gradio layouts

Add HTML and update cache

Allow overriding the overall prompt

Add notes

Update captions (remove absolute path)

Update cache path

Fix gradio version

Add a note on the environment

Update README

Add example caches

Add caching and allow scale boxes

generator seed fix for baseline

Allow using different schedulers and negative prompts

Add examples, adjustable number of steps, and custom template

Add stage 2

Update

Add demo

initial commit