Bad output

[ArpBansal]

Can you tell me the possible reason i am getting for this type of output:

doing for 20 inference step.
prompt: A cat holding a sign that says hell world, post-apocalyptic, dark-night

takes around 9 seconds
doing text to image only
with torch.no_grad():
result = pipeline(
prompt=payload.prompt,
generation_mode="image",
num_inference_steps=steps,
height=payload.height,
width=payload.width,
generator=generator,
)
dependencies:
dependencies = [
"diffusers==0.35.1",
"transformers<5",
"accelerate",
"timm",
"einops",
"sentencepiece",
"protobuf",
"pillow",
"https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu12torch2.9cxx11abiTRUE-cp312-cp312-linux_x86_64.whl"
]

[nini0919]

Hi, this is our inference result based on the prompt you provided:
"A cat holding a sign that says hell world, post-apocalyptic, dark-night"

Please note that your prompt contains "hell world" rather than "hello world". So if the generated result looks like "hello" missing the last "o", this may come from the prompt itself.
Also, the height and width in our test were set according to the aspect ratio of the image you posted.
For reference, here are the dependency versions used in our local environment:

Our inference script is as follows:

import torch
from internvlu import InternVLUPipeline

prompt = """A cat holding a sign that says hell world, post-apocalyptic, dark-night"""

pipeline = InternVLUPipeline.from_pretrained(
    "Your_checkpoint_here",
    torch_dtype=torch.bfloat16,
)

pipeline.to("cuda")

with torch.no_grad():
    image = pipeline(
        prompt=prompt,
        generation_mode="image",
        height=576,
        width=1024,
        generator=torch.Generator(device="cuda").manual_seed(42)
    ).images[0]

image.save("./example_t2i.png")