Quick fix for negative_prompt_embeds is None (#159)

* Fix unique_id comp order

* Fix LTX2 link

* Fix param size

* Fix param size

* [docs] reorder supported algorithms by arXiv date and add paper links

Improve README discoverability by sorting trainer entries by first arXiv upload time and adding direct paper references for each algorithm.

Made-with: Cursor

* [docs] add separate Flow-GRPO and Dance-GRPO paper links

Clarify the GRPO row in the Supported Algorithms table by linking Flow-GRPO and Dance-GRPO to their respective papers.

Made-with: Cursor

* [chore] update diffusers submodule to latest upstream commit (#142)

Made-with: Cursor

* Quick fix for negative_prompt_embeds is None

Author: Jayce-Ping

src/flow_factory/models/wan/wan2_v2v.py

@@ -322,7 +322,8 @@ def inference(
             negative_prompt_embeds = encoded.get("negative_prompt_embeds", None)
         else:
             prompt_embeds = prompt_embeds.to(device)
-            negative_prompt_embeds = negative_prompt_embeds.to(device)
+            if negative_prompt_embeds is not None:
+                negative_prompt_embeds = negative_prompt_embeds.to(device)
 
         batch_size = prompt_embeds.shape[0]
 

src/flow_factory/models/z_image/z_image.py

@@ -240,7 +240,8 @@ def inference(
             negative_prompt_embeds = encoded['negative_prompt_embeds'] if do_classifier_free_guidance else None
         else:
             prompt_embeds = [pe.to(device) for pe in prompt_embeds]
-            negative_prompt_embeds = [npe.to(device) for npe in negative_prompt_embeds]
+            if negative_prompt_embeds is not None:
+                negative_prompt_embeds = [npe.to(device) for npe in negative_prompt_embeds]
 
         batch_size = len(prompt_embeds)
         num_channels_latents = self.pipeline.transformer.in_channels