Alpha-VLLM / Lumina-T2X

Lumina-T2X is a unified framework for Text to Any Modality Generation
MIT License
2.04k stars 86 forks source link

🤩 [User Study] Report your bad examples! #19

Open PommesPeter opened 4 months ago

PommesPeter commented 4 months ago

Hi, all

Thank you for being so interested in Lumina-T2X. If you encounter images of poor quality, please feel free to report them in this issue to assist in our model's enhancement.

You can directly add a comment on this issue and use the template below:

prompt: <copy-paste prompt here>
image: <copy-paste image here>

You can also optionally report the hyperparameters you have used if it does influence the quality.

jrd-rocks commented 3 months ago

Lumina-Next-SFT

prompt: a giant mech reaching out to a skyscraper image: image-1

prompt: dramatic digital painting of an alluring mermaid sitting on top of a rock that emerges from the wild and roaring sea, under the waterline the base of the rock and seafloor is littered with human skulls and bones, a deceptively tranquil and peaceful atmosphere image: image

Maybe prompts like these aren't fair, but i find that, apart from the big hosted-only models, generative image models generally just can't do slightly more complex scenes like this. Describing how elements look often goes fine, but describing the composition of (objects on) a scene, often goes awry unless it's really basic.

(The gens aren't very coherent (mech head and limbs, mermaid face and tail), but that's not why i post them, I'm sure further training will help with that)

gaopengpjlab commented 3 months ago

@jrd-rocks Your feedback is greatly appreciated.

heart-du commented 3 months ago

为什么绿植感觉一直生成的不太好。 ![Uploading image.png…]()

heart-du commented 3 months ago

A garden where the shrubs are three-dimensional and layered, and every leaf and branch can be seen. 不够立体和真实

gaopengpjlab commented 3 months ago

Thanks for your feedback. We will look into the problem.

gaopengpjlab commented 3 months ago

Can you share your failure cases?