--Krea2
Actually i tested with different combinations raw and turbo. For best quality i use raw with cfg4, for fast result i use raw, but witth cfg1 and turbo lora (it really gives better quality then kreo2 turbo model). U can use with other loras freely, also works good with Lenovo lora and nsfw loras. recommended weight 0.6-0.8
--Anima
Trained on anima3. Recommended strength: 0.4.
Works best with block filtering — I created a ComfyUI nodepack for this: ComfyUI-LoRA-Block-Filter. (use no_style_leak preset). Examples generated alongside my other LoRA — Lenovo.
Experimental version. Results may vary depending on prompt and block weight settings.
--Qwen2512
Trained with the best my technique. Best use with Samsung lora
-- Z-Image (Asian Edition)
I'm experimenting with a completely different training method here. While it's still a work in progress, feel free to give it a try.
Regarding the examples: I used 40 steps to test the limits, but you can definitely achieve good results with fewer steps like 20 for example
-- Chroma
Use with weight around 0.6-0.8. Works great along with lenovo lora
-- Z-Image
Unfortunately at this moment only european girls, but this was just test, so wait for new versions. Highly recommend to use with my workflow https://civarchive.com/models/2190193?modelVersionId=2466066
-- Qwen
Alternative version gives more clean result with less noise, blur, grain. I recommend when blend with other loras to set weight around 0.6
Just a beauty girls for qwen. Use it along with my Lenovo lora for Qwen
You can take workflow here: https://huggingface.co/Danrisi/Lenovo_Qwen/resolve/main/Qwen_danrisi.json
Description
FAQ
Comments (15)
I can't get it to look quite as realistic as it does in your previews- what are your settings (cfg, scheduler, sampling method, etc)? Are you on base Anima 3.0?
I’m planning to upload the workflow later today. I’ve put together some specific tweaks to achieve that realistic look, so it’ll be much easier to show you everything in the file itself
@Danrisi That would be much appreciated! I tried using the lora on Forge Neo and the results were pretty bad, but looks like all your examples were made in comfy
Hey can somebody tell me if it works for the distilled 9B version or is it exclusively for the undistilled 9B-base?
Klein loras are all (or should be) trained on the base model. If they are, they work for both.
@Duke_Silver thanks for clarifying
Can u do for LTX 2.3 ???
Hi Danrisi,
I hope you’re doing well. I wanted to ask if it would be possible to request additional LoRA models.
I’ve been using several of your LoRAs, and I really appreciate the variety—especially for the Flux Klein versions, which already have a decent range. However, for the Qwen 2512 version, there are currently only two available (Nice Girl and Samsung Camera), which feels quite limited. I would really love to see more diversity there.
These days, I’ve been focusing more on video creation using LTX 2.3, and from my experience, Qwen 2512 is one of the most realistic models for local workflows. When combined with your LoRAs, it produces a very distinct amateur-style, early 2000s snapshot aesthetic that feels more authentic than anything else I’ve tried.
My current workflow looks like this:
Generate anchor images using Qwen 2512 + your LoRA
Refine those anchor images with Flux Klein (camera adjustments, motion, consistency)
→ to produce consistent scenes for LTX 2.3
Use the final images as multi-image inputs in LTX 2.3 with prompt relay nodes to create short, handheld-style video clips
In this process, the biggest limitation right now is the lack of diverse LoRAs for the initial “anchor images.” Because of that, it’s difficult to create varied yet consistent outputs.
Personally, I’m especially interested in LoRA styles such as low-quality / lo-fi looks, Android phone camera, iPhone camera, handheld aesthetics, and your signature Danrisi-style female character presets. I feel that having more variations in these directions would be incredibly useful for my workflow.
I also think the Flux Klein versions already have a good selection, but additional variations there would still be very valuable and appreciated.
If possible, I would greatly appreciate it if more LoRAs—especially for Qwen 2512 and Flux Klein—could be developed in the future.
Thank you again for your amazing work.
Are there any specific trigger words for the Anima Lora? Also, I find your Lora block filtering node really interesting, have you played around or have any source of what each Block and Layers can do?
ZIT and ZIB loras are not available for download, as are some other...
real shit. i'll check tomorrow and reupload them, anyway, u can download them on my HF https://huggingface.co/Danrisi
This is actually really solid for ANIMA at 1 Strength.
Props for not falsifying any of your results and including the natural artifacts. Downloaded it for both playing around with and especially the honesty.