Type | |
Stats | 1,209 |
Reviews | (253) |
Published | Mar 27, 2023 |
Base Model | |
Training | Steps: 50,000 |
Trigger Words | fantasycharacter |
Hash | AutoV2 04D1FCDBB6 |
What is this?
Style of heroic fantasy animal-human mixes. Trained with a dataset of 68 character concept art images.
How to use
Save to your embeddings folder in SD installation. Prompt the file with fantasycharacter
Recommended settings
CFG Scale: I tested all images with 7, feel free to experiment with lower values, from 4 to 5.5 to give it more freedom.
Steps: 30 will produce concept art, 50 solid digital artwork, 100-150 for highly detailed renders.
Positive prompt
Formula:
[Medium] of [Subject constants], [Subject Descriptions], [Style tags], [Beautifier tags], [Model Specific Tags]
See a complete list of positive prompts you can use in the version description here, just add your desired medium in front.
Negative prompt:
Because of a few training images, it is recommended to add these:
watermark, logo, artist name, signature, text
The example I used is a mix of a few negative embeddings plus some words I added myself to fine-tune the results:
ng_deepnegative_v1_75t, bad-artist, bad-hands-5, lowres, ((bad anatomy)), blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), logo, cropped, worst quality, low quality, normal quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts)), (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), artist name, watermark, logo, signature, text, out of the image, out of frame, pale colors, NSWF, nude
Canvas size
512x768 with a highres fix at 1.5 times and Denosing at 0.5
Model recommendations
I tested this on 10+ models (see examples) and concluded this:
Anime models will do humans with cat ears. Tested 3-4, maybe you get lucky.
Standard quality results with general purpose models like Deliberate and Midjourney family, DreamShaper and Dreamlike Diffusion.
Interesting results with specialized models like RPGv4 (killing on the outfit side), should be interesting with "furry" models (for example FurryMix) or PixarStyle for Disney-like characters.
Amazing results with WestMagic. I expect Asian graphics models to kill the outfit and standing pose.
All the models you can find in the Civitai search.
Beyond the Examples
Feel free to experiment with other models, style strengths (other than 1.0), low CFG (3-5.5) and high steps (100-150) combinations on high-performing models, or anything else that crossed your mind.
Make sure to share the results in the comments.
Follow us for more
Peaksel team is inspired by the willingness of this community to share and is grateful for all the tools and knowledge we got here. This is why we plan to share all the best tools we make. Follow us to stay in the loop.