quick workerA big move was made.Open SourceI've got my own image generation modelKetu Kolors. This is not an ordinary model; it has trained on billions of text images, carrying the Universal Language Model (GLM) as a text encoder, supporting bilingual Chinese and English language hints, and processing a context of 256 tokens。
Kolors Features at a Glance:
- Chinese and English bilingual support:The general language model (GLM) is used as the text encoder, so that the model is not only proficient in English, but also can perfectly understand and use Chinese prompts.
- Long text processing capabilities:Supporting a context length of up to 256 tokens, it allows creators to describe their ideas in detail, whether it is a complex scene or a rich story.
- Massive data training:Trained on billions of text-image pairs, the model has a large knowledge base and is able to generate diverse and accurate images.
- Optimization of Chinese cultural elements:Special optimization has been carried out for Chinese cultural elements, making the generated images more in line with Chinese cultural characteristics and meeting localization needs.
- Chinese text generation:The Graphical Kolors not only understands Chinese, but also embeds Chinese text in the resulting picture, adding more expression to the image。
After testing, I found that the performance of inserting Chinese into pictures is better now, and it can basically be output correctly, but for English, it is easy to miss words or make mistakes.
As you can see, the Chinese version of the lying cat generated above is completely fine, but when I change it to "AIbase", some characters are missing. As far as Chinese output is concerned, Ketu's performance is remarkable, but please note that the text should not be too long, otherwise it is easy to make mistakes.
This model is not just a simple tool, it has the powerful technical support of Kuaishou. It is trained on massive data and has special optimization for Chinese cultural elements, so the generated images are more Chinese. This is not only a technological breakthrough, but also a cultural inheritance.
The open source plan also includes CN (ControlNet) support, LoRa (low-rank adaptation), IPA (image prompt adaptation) and ComfyUI direct support, all of which are designed to make your creative process more smooth and personalized.
Technical details:
- “Fartable Kolors” is based on the SDXL model structure and incorporates ChatGLM256 technology to enhance bilingual understanding and text generation。
- It is worth noting that running this model requires a large amount of video memory, about 19GB, which may have certain requirements on the hardware device.
The quick-hand open source, Kolors, is not only a contribution to the technological community, but also a bold boost to creative freedom. This shows the technical resolve and strength of fast hands in AI and shows us the infinite potential of AI in artistic creation。
Ketu official website: https://top.aibase.com/tool/kuaishouketudamoxingkolors
Project address:https://www.1ai.net/12103.html