If you like my content, support me at:
https://ko-fi.com/duchaiten
Patreon is here (but there's nothing in it):
https://linktr.ee/Duc_Haiten
I created my Discord server. I'm still quite new to Discord server management, feel free to share your opinions.
https://discord.gg/vKEW6jqa49
v0.3 makes more realistic and more cosplay data
Beta_v0.2 is a version that adds a lot of anime data. It slightly suppresses the NSFW data, and the real 3D material is also affected a bit, but since it's just a beta version, it's fine. I’ll try different directions to see which one people like the most.
This model was retrained from Qwen image for a full month (counting only continuous training time). I haven’t finished it yet, but since it’s been taking way too long, I decided to release the beta version first. It seems that Qwen doesn’t work with A1111 or ForgeUI yet, so I ran it on ComfyUI instead. Anyway, I hope everyone enjoys this new model after such a long break — if you run into any issues, please let me know.
Oh, and about the sample images — there are some things I didn’t do in them, but that doesn’t mean the model can’t do those things or that it needs a LoRA. You know what I mean 😉.
(worst quality:1.5), (low quality:1.5), (normal quality:1.5), lowres, bad anatomy, bad hands, multiple eyebrow, cropped, extra limb, missing limbs, deformed hands, long neck, long body, signature, username, artist name, conjoined fingers, deformed fingers, ugly eyes, imperfect eyes, skewed eyes, unnatural face, unnatural body, error, painting by bad-artist
About Us — LahGenAI
Our team includes known creators in the AI art and workflow community such as: @Lah_Inthe_Futureland , @DucHaiten, @Hius, ChugAx from shakker and @Hideroki
Visit us at: https://lahteam.vn
Our Services
We specialize in:
Finetuning Checkpoints and LoRAs for image and video applications.
Building ComfyUI workflows for image/video processing and integrating them into AI production pipelines.
Deploying and packaging AI products, including solution consulting and product development support for startups.
If you need to build a stable, scalable visual AI pipeline, we can support your team end-to-end.
Description
FAQ
Comments (10)
Can you provide prompts and or workflows for the gallery images? Thanks for your time and effort.
i added the prompts to the image, but i'm not sure how to put the workflows up here, but actually it's the default workflows on the qwen image, i just tweaked it a little bit, i'll post the workflows image
@DucHaiten I appreciate it. It has to do with how you save the image - if you choose 'save metadata' in the Save Image node, it should keep the workflow, etc. attached.
I too am working on training a QWEN model, though I'm still working on the dataset. QWEN seems to have superior language understanding and generating capabilities when compared to Flux or IL. I know how hard training these larger models can get so here's some buzz. I've been impressed by your other models also, so keep up the good work.
thank you, i am definitely still working on it every day
thanks for your work, I give you 2K buzz
steps, sampler, cfg?
20 steps with CFG 5 works for me. QWEN lightning LoRA allows me to cut it down to 4 steps.
“Great model! Quick question: Where can I find the list of artist tags used for training? And was it trained on both natural language and Danbooru tags?”
I want to train it so that people don't care about the artist name, and yes it's trained on natural language and tags