TIPO: Text to Image with text presampling for Prompt Optimization

500M LLaMA arch model trained for TIPO.
Tech Report: https://hackmd.io/@KBlueLeaf/BJULOQBR0

Introduction

In this project, we introduce "TIPO" (Text to Image with text presampling for Prompt Optimization), an innovative framework designed to significantly enhance the quality and usability of Text-to-Image (T2I) generative models. TIPO utilizes the Large Language Models (LLMs) to perform "Text Presampling" within the inference pipeline of text-to-image generative modeling. By refining and extending user input prompts, TIPO enables generative models to produce superior results with minimal user effort, making T2I systems more accessible and effective for a wider range of users.

Usage

Use updated version of DTG extension (renamed to z-tipo-extension), current version of z-tipo-extension support stable-diffusion-webui, stable-diffusion-webui-forge and ComfyUI. SD-Next haven't been tested. https://github.com/KohakuBlueleaf/z-tipo-extension

Model arch and Training

This model is LLaMA arch with 500M parameters, the training data is combined version of Danbooru2023, GBC10M and Coyo-HD-11M.
The total token seen is around 30B tokens.
For more information please refer to the tech report and following table.

	TIPO-200M	TIPO-500M
Arch	LLaMA	LLaMA
Max ctx length	1024	1024
Batch Size	2048	3584
Training dataset	Danbooru, GBC10M, 5epoch Danbooru, GBC10M, Coyo11M, 3epoch	Danbooru, GBC10M, Coyo11M, 5epoch
Real Token Seen*	40B token	30B token
Training Hardware	RTX 3090 x 4	H100 x 8
Training Time	420 hour`	100 hour`
URL	KBlueLeaf/TIPO-200M · Hugging Face	KBlueLeaf/TIPO-500M · Hugging Face

*: We only count "non-padding token" in the token seen, since all the training data have very large length range
`: Since the training data is pretty short, it cost more time to reach same token seen than general LLM pretraining.
As reference, with 4096 as max ctx length and almost all the data have reach that length, you may only need 2days to reach 10B token seen on RTX 3090 x 4 with 200M model.

Evaluation

We have tested TIPO in several metric:

1. Aesthetic Score (Higher is Better)

We compute the Aesthetic Score using the Aesthetic Predictor V2.5. This metric is calculated on the short/truncated long test.

Figure 1: Aesthetic Score distribution.

2. AI Corrupt Score (Higher is Better)

The AI Corrupt Score is obtained from the AICorruptMetrics in sdeval.

This metric is calculated on the short/truncated long test.

Figure 2: AI Corrupt Score distribution.

3. Frechet Dino Distance (FDD) on Scenery Tag Test

We use FDD on the Scenery Tag Test to demonstrate that when input prompts address a smaller distribution, the model struggles to generate images that reflect the true distribution. However, with TIPO, this issue is mitigated.

FDD Model	`<meta> scenery` only	`<meta> scenery` + TIPO
DinoV2 ViT-S	0.1917	0.1786
DinoV2 ViT-B	0.2002	0.1755
DinoV2 ViT-L	0.2017	0.1863
DinoV2 ViT-G	0.2359	0.2096

Table 1: Frechet Dino Distance (FDD) on Scenery Tag Test.

LICENSE

This model is released under Kohaku License 1.0
You can check the above provided URL or check the LICENSE file in this repo.

Citation

@misc{yeh2024tipo,
  title = {TIPO: Text to Image with text presampling for Prompt Optimization},
  author = {Yeh, Shih-Ying},
  year = {2024},
  month = {9},
  day = {29},
  note = {Technical report available at \url{https://hackmd.io/@KBlueLeaf/BJULOQBR0}. 
          Model available at \url{https://huggingface.co/KBlueLeaf/TIPO-500M}. 
          Source code available at \url{https://github.com/KohakuBlueleaf/KGen}},
}

KBlueLeaf
/

TIPO-500M