black-forest-labs

/

FLUX.1-dev

like 4.8k

How would I create a random image?

I think it should be something like this:

prompt_embeds = pipeline._get_clip_prompt_embeds(
    "",
    device="cuda",
    num_images_per_prompt=1
)
print(f"prompt_embeds shape: {prompt_embeds.shape}")

joint_attention_dim = pipeline.transformer.config.get('joint_attention_dim', 4096)
pooled_prompt_embeds_shape = (768, joint_attention_dim)
print(f"pooled_prompt_embeds shape: {pooled_prompt_embeds_shape}")

pooled_prompt_embeds = torch.rand(pooled_prompt_embeds_shape, dtype=torch.float32).to(dtype=torch.bfloat16, device='cuda')
with torch.no_grad():
  generated_image = pipeline(prompt_embeds=prompt_embeds, pooled_prompt_embeds=pooled_prompt_embeds, generator=generator).images[0]

But there is a mismatch in shape

115
116 def forward(self, input: Tensor) -> Tensor:
--> 117 return F.linear(input, self.weight, self.bias)
118
119 def extra_repr(self) -> str:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (768x4096 and 768x3072)