CivArchive
    Cat Tower (NoobAI XL checkpoint) - v1.4 (Epsilon-pred)
    NSFW
    Preview 91943975
    Preview 91944126
    Preview 91945604
    Preview 91946572
    Preview 91946619
    Preview 91946721
    Preview 91946812
    Preview 91947584
    Preview 91949118
    Preview 91949145
    Preview 91949166
    Preview 91949244
    Preview 91949285
    Preview 91949310

    Cat Tower is anime style NoobAI XL checkpoint.

    VAE is baked in.

    v-prediction versions are experimental models.

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

    • ComfyUI (recommended)

    • Forge Classic (recommended)

    • reForge

    • Forge (maybe)

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

    License

    This model adheres to the NoobAI-XL License (Modified Fair AI Public License 1.0-SD) and is available for non-commercial use only. This license prohibits any form of commercialization, including but not limited to monetization or commercial use of the model, derivative models, or model-generated products.

    • Steps: 25-30

    • CFG scale: 5-7

    • Sampler (vPred): I recommend using "Euler". Using "Euler a" sometimes generates colored skin and objects such as red, blue and so on.

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

    • Scheduler (vPred) : 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 (ChenkinNoob v1.1)

    1. Merging the difference between ChenkinNoob-XL-V0.2 and stable-diffusion-xl-base-1.0, CyberRealistic XL v7.0 and stable-diffusion-xl-base-1.0 to ChenkinNoob-XL-V0.2 by "Perpendicular Component" and "Add Difference" (Checkpoint A)

    2. Merging Checkpoint A to ChenkinNoob-XL-V0.2 by "SLERP", alpha = 0.5 (Checkpoint B)

    3. Merging Checkpoint B to ChenkinNoob-XL-V0.2 by "Rotate", alignment = 1.0, alpha = 0.0 (Checkpoint C)

    4. Replacing the CLIP of Checkpoint C by ChenkinNoob-XL-V0.2 (Checkpoint D)

    5. Finetuning Checkpoint D with anime style dataset(0.5k) by OFT method, dim = 4, alpha = 1e-3, learning rate = 2.5e-6, 15000 steps. And merging the OFT model to ChenkinNoob-XL-V0.2, ratio = 1.0 (Checkpoint E)

    6. Merging Checkpoint E to Checkpoint D by "SLERP" with block (0,0,0,0,0,0,0,0,0,0,0,0.7,0.7,0.7,0.7,0,0,0,0,0) (Checkpoint F)

    7. Merging Cat Carrier v8.0 to Checkpoint F by "Weighted Sum" with block (0,0,0,0,0,0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0,0,0,0,0) (Checkpoint G)

    8. Merging the difference between Cat Carrier v8.0 and Raehoshi illust XL v7.1 to Checkpoint G by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0) (Checkpoint H)

    9. Merging the difference between LunarPeachMix v2.0 and Illustrious XL v1.1 to Checkpoint H by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.3,0.3,0,0,0,0,0,0,0) (Checkpoint I)

    10. Merging the difference between copycat-RouWei vpred-v0.42 and Rouwei 0.80 vpred to Checkpoint I by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.3,0.3,0,0,0,0,0,0,0) (Checkpoint J)

    11. Merging style LoRA to Checkpoint J (Cat Tower ChenkinNoob v1.1)

    Merge recipes (ChenkinNoob v1.0)

    1. Merging the difference between ChenkinNoob-XL-V0.1 and stable-diffusion-xl-base-1.0, CyberRealistic XL v7.0 and stable-diffusion-xl-base-1.0 to ChenkinNoob-XL-V0.1 by "Perpendicular Component" and "Add Difference" (Checkpoint A)

    2. Merging Checkpoint A to ChenkinNoob-XL-V0.1 by "SLERP", alpha = 0.5 (Checkpoint B)

    3. Merging Checkpoint B to ChenkinNoob-XL-V0.1 by "Rotate", alignment = 1.0, alpha = 0.0 (Checkpoint C)

    4. Replacing the CLIP of Checkpoint C by ChenkinNoob-XL-V0.1 (Checkpoint D)

    5. Finetuning Checkpoint D with anime style dataset(0.5k) by OFT method, dim = 4, alpha = 1e-3, learning rate = 2.5e-6, 15000 steps. And merging the OFT model to ChenkinNoob-XL-V0.1, ratio = 1.0 (Checkpoint E)

    6. Merging Checkpoint E to Checkpoint D by "SLERP" with block (0,0,0,0,0,0,0,0,0,0,0,0.7,0.7,0.7,0.7,0,0,0,0,0) (Checkpoint F)

    7. Merging Cat Carrier v8.0 to Checkpoint F by "Weighted Sum" with block (0,0,0,0,0,0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0,0,0,0,0) (Checkpoint G)

    8. Merging the difference between Cat Carrier v8.0 and Raehoshi illust XL v7.1 to Checkpoint G by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0) (Checkpoint H)

    9. Merging the difference between LunarPeachMix v2.0 and Illustrious XL v1.1 to Checkpoint H by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.3,0.3,0,0,0,0,0,0,0) (Checkpoint I)

    10. Merging the difference between copycat-RouWei vpred-v0.42 and Rouwei 0.80 vpred to Checkpoint I by "Train Difference" with block (0,0,0,0,0,0,0,0,0,0,0,0.3,0.3,0,0,0,0,0,0,0) (Checkpoint J)

    11. Merging style LoRA to Checkpoint J (Cat Tower ChenkinNoob v1.0)

    Merge recipes (Obsession vPred v2.0)

    1. Merging the difference between Obsession (Illustrious-XL) v-pred_v2.0 and stable-diffusion-xl-base-1.0, CyberRealistic XL v5.6 and stable-diffusion-xl-base-1.0 to Obsession (Illustrious-XL) v-pred_v2.0 by "Perpendicular Component" and "Add Difference" (Checkpoint A)

    2. Merging Checkpoint A to Obsession (Illustrious-XL) v-pred_v2.0 by "SLERP", alpha = 0.5 (Checkpoint B)

    3. Merging Checkpoint B to Obsession (Illustrious-XL) v-pred_v2.0 by "Rotate", alignment = 1.0, alpha = 0.0 (Checkpoint C)

    4. Replacing the CLIP of Checkpoint C by Obsession (Illustrious-XL) v-pred_v2.0 (Checkpoint D)

    5. Finetuning Checkpoint D with anime style dataset A(2.1k) by full finetune, optimizer = AdamW8bit, scheduler = warmup stable decay, learning rate = 6e-6, scheduler min lr ratio = 0.1, 32 epoch, unet only, gradient accumulation steps = 1, batch size = 4. (Checkpoint E)

    6. Finetuning Checkpoint D with anime style dataset B(0.5k) by full finetune, optimizer = AdamW8bit, scheduler = warmup stable decay, learning rate = 3e-6, scheduler min lr ratio = 0.1, 60 epoch, unet only, gradient accumulation steps = 1, batch size = 4. (Checkpoint F)

    7. Merging Checkpoint F to Checkpoint E by "SLERP", alpha = 0.5 (Checkpoint G)

    8. Merging the difference between LunarPeachMix v2.0 and Illustrious XL v1.1 to Checkpoint G by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint H)

    9. Merging the difference between JANKU v5.0 and RouWei v0.8 epsilon to Checkpoint H by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint I)

    10. Merging the difference between copycat-RouWei vpred-v0.42 and Rouwei 0.80 vpred to Checkpoint I by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint J)

    11. Merging the Checkpoint I by "Clamp", bounds = Checkpoint G, LunarPeachMix v2.0, JANKU v5.0 and copycat-RouWei vpred-v0.42 (Checkpoint K)

    12. Add v_pred and ztsnr keys to Checkpoint K by noob_v_pencil-XL merge recipes (Cat Tower vPred v2.0)

    Merge recipes (Epsilon-pred v1.4)

    1. Merging the difference between NoobAI-XL Epsilon-pred v1.1 and stable-diffusion-xl-base-1.0, CyberRealistic XL v5.6 and stable-diffusion-xl-base-1.0 to NoobAI-XL Epsilon-pred v1.1 by "Perpendicular Component" and "Add Difference" (Checkpoint A)

    2. Merging Checkpoint A to NoobAI-XL Epsilon-pred v1.1 by "SLERP", alpha = 0.5 (Checkpoint B)

    3. Merging Checkpoint B to NoobAI-XL Epsilon-pred v1.1 by "Rotate", alignment = 1.0, alpha = 0.0 (Checkpoint C)

    4. Replacing the CLIP of Checkpoint C by NoobAI-XL Epsilon-pred v1.1 (Checkpoint D)

    5. Finetuning Checkpoint D with anime style dataset(0.5k) by OFT method, dim = 4, alpha = 1e-3, learning rate = 2.5e-6, 15000 steps. And merging the OFT model to Checkpoint D, ratio = 1.0 (Checkpoint E)

    6. Merging Cat Carrier v6.0 to Checkpoint E by "Weighted Sum" with block (0,0,0,0,0,0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0,0,0,0,0) (Checkpoint F)

    7. Merging the difference between LunarPeachMix v2.0 and Illustrious XL v1.1 to Checkpoint F by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint G)

    8. Merging the difference between JANKU v4.0 and RouWei v0.7 epred to Checkpoint G by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint H)

    9. Merging the difference between copycat-RouWei vpred-v0.42 and Rouwei 0.80 vpred to Checkpoint H by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint I)

    10. Merging the Checkpoint I by "Clamp", bounds = Checkpoint F, LunarPeachMix v2.0, JANKU v4.0 and copycat-RouWei vpred-v0.42 (Cat Tower Epsilon-pred v1.4)

    Merge recipes (vPred v1.8)

    1. Merging the difference between NoobAI-V-Pred-1.0-Version and stable-diffusion-xl-base-1.0, CyberRealistic XL v5.6 and stable-diffusion-xl-base-1.0 to NoobAI-V-Pred-1.0-Version by "Perpendicular Component" and "Add Difference" (Checkpoint A)

    2. Merging Checkpoint A to NoobAI-V-Pred-1.0-Version by "SLERP", alpha = 0.5 (Checkpoint B)

    3. Merging Checkpoint B to NoobAI-V-Pred-1.0-Version by "Rotate", alignment = 1.0, alpha = 0.0 (Checkpoint C)

    4. Replacing the CLIP of Checkpoint C by NoobAI-V-Pred-1.0-Version (Checkpoint D)

    5. Finetuning Checkpoint D with anime style dataset(0.5k) by OFT method, dim = 4, alpha = 1e-3, learning rate = 2.5e-6, 15000 steps. And merging the OFT model to Checkpoint D, ratio = 1.0 (Checkpoint E)

    6. Merging Cat Carrier v6.0 to Checkpoint E by "Weighted Sum" with block (0,0,0,0,0,0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0,0,0,0,0) (Checkpoint F)

    7. Merging the difference between LunarPeachMix v2.0 and Illustrious XL v1.1 to Checkpoint F by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint G)

    8. Merging the difference between JANKU v4.0 and RouWei v0.7 epred to Checkpoint G by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint H)

    9. Merging the difference between copycat-RouWei vpred-v0.42 and Rouwei 0.80 vpred to Checkpoint H by "Add Difference" with block (0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) (Checkpoint I)

    10. Merging the Checkpoint I by "Clamp", bounds = Checkpoint F, LunarPeachMix v2.0, JANKU v4.0 and copycat-RouWei vpred-v0.42 (Checkpoint J)

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

    Merge recipes (Epsilon-pred v1.3)

    1. Merging Cat Tower Epsilon-pred v1.2 and Cat Tower v-pred v1.7 by "DARE-TIES" (Cat Tower Epsilon-pred v1.3)

    Merge recipes (vPred v1.7)

    1. Merging the difference between CyberRealistic XL v5 and stable-diffusion-xl-base-1.0 to NoobAI-V-Pred-1.0-Version by "Add Difference" (Checkpoint A)

    2. Replacing the CLIP of Checkpoint A by NoobAI-V-Pred-1.0-Version (Checkpoint B)

    3. Merging Cat Carrier v3.0 to Checkpoint B by "Weighted Sum" with block (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.4,0.4,0.4,0.4,0.4,0.4,0.0,0,0,0,0) (Checkpoint C)

    4. Merging Cat Tower vpred v1.5 to Checkpoint C by "Add Cosine B" 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) (Checkpoint D)

    5. Merging Cat Bread eps-pred v1.0 to Checkpoint D 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) (Checkpoint E)

    6. Merging the difference between Catloaf and ponyDiffusionV6XL to Checkpoint E 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 F)

    7. Merging the difference between copycat-noob Vpred_v1.01 and NoobAI-V-Pred-1.0-Version to Checkpoint F 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 G)

    8. Merging style LoRA to Checkpoint G (Checkpoint H)

    9. Add v_pred and ztsnr keys to Checkpoint H by noob_v_pencil-XL merge recipes (Cat Tower vPred v1.7)

    Merge recipes (vPred v1.6)

    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 Checkpoiint B to Checkpoint A by "Weighted Sum", alpha = 0.4 (Checkpoint C)

    5. Merging Cat Carrier v3.0 to Checkpoint C by "Weighted Sum" with block (0,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.6,0.6,0.6,0.4,0.6,0.6,0.0,0,0,0,0) (Checkpoint D)

    6. Merging Checkpoint D and HarmoniqMix_vPred_v1 by "Weighted Sum" 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 Catloaf and ponyDiffusionV6XL to Checkpoint E 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 F)

    8. Merging the difference between copycat-noob Vpred_v1.01 and NoobAI-V-Pred-1.0-Version to Checkpoint F 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 G)

    9. Merging style LoRA to Checkpoint G (Checkpoint H)

    10. Add v_pred and ztsnr keys to Checkpoint H by noob_v_pencil-XL merge recipes (Cat Tower vPred v1.6)

    Merge recipes (Epsilon-pred v1.2)

    1. Merging the difference between cyberrealisticXL_v31 and stable-diffusion-xl-base-1.0 to NoobAI-XL Epsilon-pred v1.1 by "Add Difference", alpha=1.0. (Checkpoint A)

    2. Replacing the CLIP of Checkpoint A by NoobAI-XL Epsilon-pred v1.1. (Checkpoint B)

    3. Merging the difference between Cat Carrier v3.0 and Checkpoint B by "Add Difference" with blocks (0.01,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.2,0.35,0.35,0.35,0.2,0.35,0.35,0,0,0,0,0). (Checkpoint C)

    4. Merging the difference between copycat-noob v0.3test and NoobAI-XL Epsilon-pred v0.75 to Checkpoint C by "Train Difference" 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 D)

    5. Merging style LoRA to Checkpoint D (Cat Tower Epsilon-pred v1.2)

    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)

    Description

    • Based on NoobAI Epsilon-pred 1.1-Version

    • The art style has been changed (like sharp and clear).

    FAQ

    Comments (12)

    DOLPH1NARUAug 2, 2025· 3 reactions
    CivitAI

    What does mean "epsilon"?

    clueless_engineerAug 2, 2025· 1 reaction

    If you've been using illustrious, then you've been using epsilon (or e-pred) models. I think only some NoobAI models are v-pred.

    nuko_masshigura
    Author
    Aug 3, 2025· 1 reaction

    @DOLPH1NARU "Epsilon-Prediction" is one of the image generation algorithms. Some models, such as NoobAI, also use an image generation algorithm called "v-Prediction."

    DOLPH1NARUAug 3, 2025

    clueless_engineer Ye, but in description Base Model NoobAI, it turns out that there are mixed algorithms here if we use the epsilon version?

    sansmiaAug 3, 2025· 1 reaction

    DOLPH1NARU NoobAI has both an e-pred and a v-pred model.

    DOLPH1NARUAug 4, 2025

    sansmia ehhhh okay

    DOLPH1NARUAug 29, 2025

    @nuko_masshigura forget to ask, that model support some artists style?

    nuko_masshigura
    Author
    Aug 29, 2025· 1 reaction

    @DOLPH1NARU Yes, you can use the same style tags as NoobAI.

    suyou3244438Aug 9, 2025· 8 reactions
    CivitAI

    画背景的神！v1.8版本是我见过的背景和场景完成度最高的noob模型,许多复杂的场景概念它都能理解且物品排序没有崩坏，太难以置信了!
    God of background painting! Version 1.8 is the noobai with the highest degree of completion in backgrounds and scenes I've ever seen. It can understand many complex scene concepts and the arrangement of items is not messed up. It's incredible!

    NTR_BLACKAug 12, 2025· 6 reactions
    CivitAI

    v1.8 is the most suitable model for Japanese anime characters I've tried. The background, style, lighting, and even the multiplayer concept are excellent.

    Although the multi-role part cannot be compared with NAI4.5, it is good enough.

    However, I'd like to ask, if a character isn't already in the model, can it be added through LoRa training? Or can I only tweak the existing character or replace its trigger word?Because the characters I added through my own training cannot work like the trigger words built into the model

    nuko_masshigura
    Author
    Aug 13, 2025· 1 reaction

    This model can also use LoRA like other models.

    Here's a LoRA I recently maked. Some of the LoRA sample images were created in cat tower vpred v1.8.

    https://civitai.com/models/1850129/

    https://civitai.com/models/1854626/

    I used IllustriousXL v0.1 as pretrained model to make these LoRA.

    If you use cat tower v-pred v1.8 as a LoRA pretrained model, do you enable the "v_parameterization" and "zero_terminal_snr" options in sd-scripts?

    The "v_parameterization" and "zero_terminal_snr" options are required when training using a v-pred model.

    NTR_BLACKAug 24, 2025

    @nuko_masshigura I have always used V1.8 as the base training, which is better. In fact, I have trained many usable VPRED LORAs. Thank you for sharing.

    Checkpoint
    NoobAI
    by nuko_masshigura
    Download (Beta) View on CivitAI
    base model
    anime

    Details

    Downloads
    2,496
    Platform
    CivitAI
    Platform Status
    Available
    Created
    8/2/2025
    Updated
    5/12/2026
    Deleted
    -

    Files

    catTowerNoobaiXL_v14EpsilonPred.safetensors

    Size:
    6.46 GB
    SHA256:
    bb68bd2706b50130b7568016512629e5e1225652753d7110dd5c8ea5187a9009

    Mirrors

    CivitAI (1 mirrors)
    TensorArt (1 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.

    SeaArt
    Cat Tower (NoobAI XL checkpoint) - v1.4 (Epsilon-pred)