Why did the authors choose such a low token max_sequence_count for this model? 128 doesn't cut it. S/B at least 512 tokens. What gives?
The token limit is a severe handicap for producing good images:
WARNING:HiDream.hi_diffusers.pipelines.hidream_image.pipeline_hidream_image:The following part of your input was truncated because max_sequence_length is set to 128 tokens: [' warm, yellow street lamps that cast a gentle, inviting light onto the cobblestones and the surrounding foliage. On the left side of the image, a large tree with bright red leaves extends its branches over the path, creating a canopy that frames the scene beautifully. The leaves on the ground contribute to the autumnal atmosphere, enhancing the sense of tranquility and natural beauty. The overall mood of the image is peaceful and nostalgic, evoking a sense of calm and timelessness. The combination of the warm lighting, the rich colors of the leaves, and the quaint architecture of the village creates a visually captivating and serene scene']