Overview
The goal of this LoRA for Flux.1 is to capture the style of Cartoon Saloon, a talented animation studio from Ireland, well-known for its unique artistic style and storytelling. Their work is all about hand-drawn 2D animation with a lovely, painterly touch, heavily influenced by Irish folklore and mythology. You'll often see richly detailed backgrounds inspired by Celtic art and design, and their stories have a way of blending the fantastical with deep emotional themes. They've a knack for weaving in messages about preserving the environment and culture, with characters that might look simple but are full of expression. Some of Cartoon Saloon's best-known films include "The Secret of Kells", "Song of the Sea" and "Wolfwalkers". These films are cherished for their stunning visuals, cultural richness, and the way they connect with both young and old. In a world full of 3D computer animation, Cartoon Saloon's style stands out for its hands-on, traditional approach, giving their work a timeless and innovative feel.
Usage
All title images were generated with the following recommended parameters:
Prompts: started with "In style of Cartoon Saloon."
Model: flux1-dev (fp8e4m3fn)
Text Encoder: t5pxxl_fp16
Sampler: euler
Scheduler: 24 steps (simple)
Flux Guidance: 4
LoRA Strength: 1Training
LoRA was created with the ostris/ai-toolkit on a single RTX 3090. Total number of images used for fine-tuning: 533. All images were captioned with Joy Caption Pre Alpha, captions were prefixed with "In style of Cartoon Saloon. " string. Specific heroes or individual film titles were not tagged. Training hyperparameters:
Rank/alpha: 16/16
Batch Size: 1
Steps: 7000
Learning Rate: 1e-4
Optimizer: adamw8bitDescription
FAQ
Comments (5)
Great lora from your own images, however for some reason it doesn't seem to influence my images much. Could be just a me problem, but I need to put in insane weight (40+) to see some variation, and it's nowhere near the style posted. Trigger words are there, so I'm not sure what's wrong.
Do you use Forge? Or NF4 checkpoint?
I've used forge for this, with a fp8 checkpoint. This worked well with all other loras I've tried (maybe 15 so far), but it's the first time it doesn't work for me here. What about you, what checkpoint / UI?
@SwahΒ I am using ComfyUI and the original fp16 checkpoint. I never used Forge, but I've heard few days ago it had problems with using some LoRAs. Do you have latest version of Forge?
I might try to test it in Forge but tbh I yet have no idea how to solve this issue if it really turns this LoRA is not working with Forge π I just used default ai-toolkit training process and did not tweak anything substantial.
@seruva19Β Right - well I certainly appreciate the replies, thanks. I actually figured it out - I needed to update my version of Forge. It mentioned "Patching LoRAs for KModel" before using the lora properly. I hope this is useful for others as well, and thanks for the quick replies!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.



















