Cat Tower (NoobAI XL checkpoint)

Cat Tower is anime style NoobAI XL checkpoint.

VAE is baked in.

If you are using Epsilon-pred v1.1 downloaded before Dec. 5th, 2024 0:00 (UTC) and only black or gray images are generated, please re-download.

v-prediction versions are experimental models.

You need to use the webui that support v-prediction.

• ComfyUI

• reForge

• AUTOMATIC1111 (from v1.4, and place the config file in the same folder as the checkpoint)

Recommended Setting

• Steps: 25-30

• CFG scale: 5-7

• Sampler (vPred v1.1- v1.5): Euler, Euler a

• Sampler (vPred v1): Euler

• Sampler (Epsilon-pred): DPM++ 2M Karras, Euler a

• Scheduler (vPred v1.1 - v1.5) : Normal, Simple ("Automatic" produce noisy images.)

• Use ADetailer as needed

Positive Prompt

masterpiece, best quality

Negative Prompt

worst quality, bad quality, low quality, lowres, scan artifacts, jpeg artifacts, sketch, light particles

Merge recipes (vPred v1.5)

1. Merging NoobAI-V-Pred-1.0-Version and IterComp by Noob Vpred itercomp merge's recipes. (Checkpoint A)

2. Finetuning NoobAI-V-Pred-1.0-Version with anime style dataset(0.5k) by OFT method, dim = 8, alpha = 1e-3, learning rate = 1e-4, 15000 steps

3. Merging the OFT model to Checkpoint A, ratio = 1.0 (Checkpoint B)

4. Merging Checkpoint B and Cat Tower v1.1 Epsilon-pred by "Weighted Sum" with block (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.4,0.4,0.4,0.4,0.0,0,0,0,0) (Checkpoint C)

5. Merging Checkpoint C and HarmoniqMix_vPred_v1 by "Weighted Sum" with block (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.0,0,0,0,0) (Checkpoint D)

6. Merging the difference between Catloaf and ponyDiffusionV6XL to Checkpoint D by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.2,0.2,0.2,0.2,0,0,0,0,0) (Checkpoint E)

7. Merging the difference between copycat-noob Vpred_v0.4 and Noobai-XL-V-pred-0.75S-Version to Checkpoint E by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.15,0.15,0.15,0.15,0,0,0,0,0) (Checkpoint F)

8. Add v_pred and ztsnr keys to Checkpoint F by noob_v_pencil-XL merge recipes (Cat Tower vPred v1.5)

Merge recipes (vPred v1.4)

1. Merging NoobAI-V-Pred-0.75S-Version and IterComp by Noob Vpred itercomp merge's recipes. (Checkpoint A)

2. Finetuning NoobAI-V-Pred-0.75S-Version with anime style dataset(0.5k) by OFT method, dim = 8, alpha = 1e-3, learning rate = 1e-4, 15000 steps

3. Merging the OFT model to Checkpoint A, ratio = 1.0 (Checkpoint B)

4. Merging Checkpoint B and Cat Tower v1.1 Epsilon-pred by "Weighted Sum" with block (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.4,0.4,0.4,0.4,0.0,0,0,0,0) (Checkpoint C)

5. Merging Checkpoint C and HarmoniqMix_vPred_v1 by "Weighted Sum" with block (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.0,0,0,0,0) (Checkpoint D)

6. Merging the difference between Catloaf and ponyDiffusionV6XL to Checkpoint D by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.2,0.2,0.2,0.2,0,0,0,0,0) (Checkpoint E)

7. Merging the difference between copycat-noob v0.3test and NoobAI-XL Epsilon-pred 0.75-Version to Checkpoint E by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.15,0.15,0.15,0.15,0,0,0,0,0) (Checkpoint F)

8. Merging NoobAI-V-Pred-0.75S-Version and Checkpoint F by "Weighted Sum" with block (0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1) (Cat Tower vPred v1.4)

Merge recipes (Epsilon-pred v1.1)

1. Merging NoobAI-XL Epsilon-pred 1.1-Version and cyberrealisticXL_v31, and replace CLIP by NoobAI-XL Epsilon-pred 1.1-Version. It's a same method of NoobaiCyberFix. (Checkpoint A)

2. Merging the difference between Cat Carrier v1.0 and Raehoshi illust XL v1.0 to Checkpoint A by "trainDifference" with blocks (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.4,0.2,0.2,0.4,0.6,0.6,0.6,0.2,0.2,0.2,0.2,0.2) (Checkpoint B)

3. Merging the difference between copycat-noob v0.3test and NoobAI-XL Epsilon-pred 0.75-Version to Checkpoint B by "trainDifference" with blocks (0,0,0,0,0,0,0,0,0.2,0,0,0.2,0.2,0.2,0.2,0,0,0,0,0) (Checkpoint C)

4. Merging style LoRA to Checkpoint C (Cat Tower Epsilon-pred v1.1)

Merge recipes (vPred v1.3)

1. Merging NoobAI-V-Pred-0.65S-Version and IterComp by Noob Vpred itercomp merge's recipes. (Checkpoint A)

2. Finetuning NoobAI-V-Pred-0.65S-Version with anime style dataset(0.5k) by OFT method, dim = 8, alpha = 1e-3, learning rate = 1e-4, 15000 steps

3. Merging the OFT model to Checkpoint A, ratio = 1.0 (Checkpoint B)

4. Merging Checkpoint B and HarmoniqMix_vPred_v1 by "Weighted Sum" with block (0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.5,0.5,0.5,0.5,0.0,0,0,0,0) (Checkpoint C)

5. Merging the difference between Catloaf and ponyDiffusionV6XL to Checkpoint C by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0.2,0.2,0.2,0.2,0,0,0,0,0) (Checkpoint D)

6. Merging the difference between copycat-noob v0.3test and NoobAI-XL Epsilon-pred 0.75-Version to Checkpoint D by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0.15,0.15,0.15,0.15,0,0,0,0,0) (Checkpoint E)

7. Merging the difference between ExilluSPO Anime v1 and illustriousXL_smoothftSPO to Checkpoint D by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0.15,0.15,0.15,0.15,0,0,0,0,0) (Checkpoint F)

8. Merging NoobAI-V-Pred-0.65S-Version + (Checkpoint F - NoobAI-V-Pred-0.65S-Version) (Cat Tower vPred v1.3)

Merge recipes (vPred v1.2)

1. Merging NoobAI-V-Pred-0.65S-Version and cyberrealisticXL_v31, and replace CLIP by NoobAI-V-Pred-0.65S-Version. It's a same method of NoobaiCyberFix. (Checkpoint A)

2. Merging NoobAI-V-Pred-0.65S-Version and IterComp by Noob Vpred itercomp merge's recipes. (Checkpoint B)

3. Merging Checkpoint A * (1 - 0.5) + Checkpoint B * 0.5 (Checkpoint C)

4. Merging the difference between Cat Carrier v1.0 and Raehoshi illust XL v1.0 to Checkpoint C by "Train Difference" with block (0,0,0,0,0,0,0,0,0.25,0,0,0.25,0.45,0.45,0.45,0,0,0,0,0) (Checkpoint D)

5. Merging the difference between copycat-noob v0.3test and NoobAI-XL Epsilon-pred 0.75-Version to Checkpoint D by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.2,0.2,0.2,0.2,0,0,0,0,0) (Checkpoint E)

6. Merging style LoRA to Checkpoint E (Checkpoint F)

7. Merging NoobAI-V-Pred-0.65S-Version + (Checkpoint F - NoobAI-V-Pred-0.65S-Version) with adjust (0,0,0,0,0,2.34,0,-0.95). (Cat Tower vPred v1.2)

Merge recipes (vPred v1.1)

1. Merging NoobAI-V-Pred-0.6-Version and cyberrealisticXL_v31, and replace CLIP by NoobAI-V-Pred-0.6-Version. It's a same method of NoobaiCyberFix. (Checkpoint A)

2. Merging the difference between Cat Carrier v1.0 and Raehoshi illust XL v1.0 to Checkpoint A by "Train Difference" with block (0,0,0,0,0,0,0,0,0.3,0,0,0.3,0.5,0.5,0.5,0,0,0,0,0) (Checkpoint B)

3. Merging the difference between copycat-noob v0.3test and NoobAI-XL Epsilon-pred 0.75-Version to Checkpoint B by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.2,0.2,0.2,0.2,0,0,0,0,0) (Checkpoint C)

4. Merging style LoRA to Checkpoint C (Checkpoint D)

5. Merging NoobAI-V-Pred-0.6-Version and Checkpoint D by the following calc, NoobAI-V-Pred-0.6-Version + (Checkpoint D - NoobAI-V-Pred-0.6-Version). (Cat Tower vPred v1.1)

Merge recipes (vPred v1)

1. Merging NoobAI-V-Pred-0.5-Version and cyberrealisticXL_v31, and replace CLIP by NoobAI-V-Pred-0.5-Version. It's a same method of NoobaiCyberFix. (Checkpoint A)

2. Merging the subtract (HarmoniqMix_vPred_v1 - NoobAI-V-Pred-0.5-Version) to Checkpoint A by "Add Difference" and "Clamp" (Checkpoint B)

3. Merging the difference between Cat Carrier v1.0 and Raehoshi illust XL v1.0 to Checkpoint B by "Train Difference" with block (0.1,0.1,0.1,0.1,0.1,0.1,0.1,0.2,0.1,0.1,0.2,0.3,0.3,0.3,0.1,0.1,0.1,0.1,0.1) (Checkpoint C)

4. Merging NoobAI-V-Pred-0.5-Version and Checkpoint C by the following calc, NoobAI-V-Pred-0.5-Version + (Checkpoint C - NoobAI-V-Pred-0.5-Version). (Cat Tower vPred v1)

Merge recipes (Epsilon-pred v1)

1. Merging the difference between Cat Carrier v1.0 and Raehoshi illust XL v1.0 to NoobaiCyberFix v1.0, NoobaiCyberFix + (Cat Carrier - Raehoshi illust XL) x alpha (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.4,0.2,0.2,0.4,0.6,0.6,0.6,0.2,0.2,0.2,0.2,0.2), mode = Add difference, calcmode = trainDifference (Checkpoint A)

2. Merging the difference between copycat-noob v0.3test and NoobAI-XL Epsilon-pred 0.75-Version to Checkpoint A, Checkpoint A + (copycat-noob - NoobAI-XL) x alpha (0,0,0,0,0,0,0,0,0.2,0,0,0.2,0.2,0.2,0.2,0,0,0,0,0), mode = Add difference, calcmode = trainDifference (Checkpoint B)

3. Merging style LoRA to Checkpoint B (Cat Tower Epsilon-pred v1)

License

This model adheres to the Fair AI Public License 1.0-SD and is available for non-commercial use only. Commercial use is prohibited.

Hide