Four Anima 2.9B Derivatives on M1 Max With a Character LoRA Remapped to 40 Blocks, All Kept the Hairstyles but Took About 1.4x as Long as Base
Contents
While putting together the October 2026 list of the latest Anima derivative checkpoints, I found that four of the derivatives from my June list now have 2.9B versions: Nova Anime AM, CottonAnima, AMIXA, and JANIMA.
My character LoRAs (small adapter weights that make a model draw specific characters) are trained on Anima-Base v1.0, whose DiT (the diffusion transformer that draws the image) has 28 blocks. The 2.9B models have 40.
On plain Anima-2.9B, the characters came out after I remapped the LoRA, that is, rewrote the block numbers in its key names to match the 40-block layout. The question was whether they would still come out on the derivative 2.9B models.
I generated images with the four latest versions on an M1 Max 64GB in ComfyUI, using the same prompts, seed, and LoRA as in the Turbo v1.1 and Aesthetic v1.1 comparison.
The seven 2B versions from the same list are covered in a separate article on the 2B versions.
Models and Basic Information
I tested the four models in the list’s table of “current versions (2.9B) of the derivatives from the June list.”
Only Nova downloaded through the Civitai download API without logging in. The other three returned HTTP 401 (authentication required) without an API key.
CottonAnima 2.9B_v1 has a mirror of the same file on Hugging Face (diffusionmodels1254ani/cottonanima_29BV1_bf16). Its SHA256 hash matched the value listed on Civitai, meaning it is the same file byte for byte, so I downloaded it from there. I downloaded AMIXA2.9B and JANIMA from Civitai in a browser and checked their SHA256 hashes the same way.
| Model | Version | Release Date (UTC) | Size | File | Author’s Recommended Settings |
|---|---|---|---|---|---|
| Nova Anime AM | v6.0:2.9B | October 9 | 2.9B (40 blocks) | bf16, 5.84GB | Euler a, Normal, 20 to 30 steps, CFG 4 to 6 |
| CottonAnima | 2.9B_v1 | October 15 | 2.9B (40 blocks) | bf16, 5.84GB | er_sde, 24 to 30 steps, CFG 5 |
| AMIXA2.9B | 2.9B_V01 | October 12 | 2.9B (40 blocks) | fp16 (DiT only), 5.84GB (listed as 5.44GB by the author) | DPM++ 2M + Beta, 23 steps, CFG 4 |
| JANIMA | v1.0 2.9B | September 8 | 2.9B (40 blocks) | bf16, 5.84GB | Same as the 2B version |
Here are the parts of each author’s description that relate to these results.
| Model | Author’s Description |
|---|---|
| Nova Anime AM v6.0 | Built from Anima Base v1.0 + Aesthetic v1.1 + 2.9B. Improved prompt and context handling |
| CottonAnima 2.9B_v1 | Anima-2.9B merged with a style LoRA trained on 160 AI-generated images. Diversity drops in exchange for a fixed style. The author says to install the ComfyUI-Anima-2.9B custom node |
| AMIXA2.9B 2.9B_V01 | A merge of Anima-2.9B. Uses the coloring of the Illustrious family (SDXL-based anime models) and sharpens character details. A bf16 version bundling the VAE and text encoder is also available |
| JANIMA v1.0 2.9B | Fine-tuned from anima-base-v1.0. For the 2.9B version, the author says the new 2.9B layers were merged in to keep 2.9B’s knowledge |
Baseline and Generation Settings
The baseline is Anima-Base v1.0 at 30 steps. I reused the images from the Turbo v1.1 and Aesthetic v1.1 comparison and did not regenerate them.
| Item | The Four 2.9B Models | Base (Baseline) |
|---|---|---|
| Steps | 30 | Same as left |
| CFG | 4.5 | Same as left |
| Sampler / Scheduler | er_sde / simple | Same as left |
| Resolution | 832×1216 (portraits 832×1024, two characters 1216×832) | Same as left |
| Seed | 42 | Same as left |
| Character LoRA | keikana-animabase-v2_epoch140 with block numbers remapped for 40 blocks (strength_model 1.0, strength_clip 1.0) | keikana-animabase-v2_epoch140 (not remapped) |
| Positive Prefix | masterpiece, best quality, safe | Same as left |
Some of these models recommend Euler a or DPM++ 2M, but I kept er_sde for all of them so the results are comparable.
| Environment | Details |
|---|---|
| Machine | M1 Max 64GB |
| ComfyUI | Set up a separate v0.33.3 instance (a version that loads 40 blocks natively). The Base images come from the October 21 article, so the ComfyUI version differs |
When I opened the Nova file, the DiT had 40 blocks, numbered 0 to 39.
The character LoRA was trained for Anima-Base’s 28 blocks, so I used the remapping script from the article on applying existing character LoRAs to Anima-2.9B to shift the block numbers to their positions in the 40-block layout. Of the 948 keys, 840 were remapped, and the 108 keys for the text-side adapter were left as they were.
ComfyUI’s log showed no warnings about LoRA keys that failed to load.
CottonAnima, AMIXA2.9B, and JANIMA also have 40 blocks numbered 0 to 39, and the same remapped LoRA loaded on them without warnings. For AMIXA2.9B, I used the fp16 DiT-only file instead of the bf16 all-in-one file. Anima uses Qwen3 0.6B as its text encoder. JANIMA’s page also distributes one, but its SHA256 matched my local qwen_3_06b_base.safetensors, so I used the same one as the other models. I did not install the custom node the CottonAnima author asks for and loaded every model with the standard loader in v0.33.3, so CottonAnima was not run in its author’s recommended setup.
Default Style
I generated a blonde character in a white robe without the character LoRA.
Test 1 (Standing Pose on a White Background)
| Model | What the Image Shows |
|---|---|
| Base | A front panel and hem in strongly yellow gold. Blunt bangs and flat shading |
| Nova Anime AM v6.0 | Hair pulled back off the forehead, embroidery on the gold front panel. Symmetrical standing pose |
| CottonAnima 2.9B_v1 | Long blonde hair off the forehead. A cape and gold sash, with the hem flaring backward |
| AMIXA2.9B 2.9B_V01 | Long blonde hair off the forehead. A wide gold front panel in the center, with gold embroidery on the sleeves and hem. Symmetrical |
| JANIMA v1.0 2.9B | Long blonde hair off the forehead. A wide gold front panel in the center, with gold embroidery on the sleeves and hem. Symmetrical |
All four drew the hair off the forehead.
AMIXA2.9B and JANIMA look alike down to the wide gold front panel, the gold embroidery on the sleeves and hem, and the symmetrical pose.
Prompt
masterpiece, best quality, safe, 1girl, solo, long blonde hair, blue eyes, white robe, gold embroidery, capelet, gold sash, long sleeves, long dress, standing, looking at viewer, full body, white backgroundNegative
worst quality, low quality, artist name, blurry, jpeg artifacts, chromatic aberrationTest 2 (Running Through a Meadow)
| Model | What the Image Shows |
|---|---|
| Base | The cape lining turns red. The castle is small and far away |
| Nova Anime AM v6.0 | A castle with a blue roof in the back left, white flowers in the foreground. Running sideways |
| CottonAnima 2.9B_v1 | A castle with a red roof on the left, sunset clouds. Running toward the viewer |
| AMIXA2.9B 2.9B_V01 | A small white castle in the back right, pink and navy clouds, white flowers in the foreground. Closer to walking than running |
| JANIMA v1.0 2.9B | Leans semi-realistic. A castle in the back right, sunset with dark clouds |
In Test 2, JANIMA switched to semi-realistic shading, a different style from its Test 1 image.
Prompt
masterpiece, best quality, safe, 1girl, solo, long blonde hair, blue eyes, white robe, gold embroidery, capelet, gold sash, long sleeves, long dress, running, wind, hair blowing, dynamic pose, fantasy landscape, castle in background, sunset sky, dramatic clouds, grass fieldThe negative prompt is the same as Test 1.
Character LoRA
I applied the character LoRA trained on Anima-Base.
Kana-chan’s identifying features are brown hair, an ahoge (a single strand of hair sticking up), a side ponytail on the viewer’s right, a blue scrunchie, brown eyes, and a white shirt with a red tie.
Kei-chan’s features are long blonde hair, blunt bangs, side braids, a blue hair ribbon in the back, blue eyes, and a red ribbon tie at the neck. She has two ribbons, and only the hair ribbon is blue.
Kana-chan
I generated a full-body standing pose.
For the portraits, I changed the prompt to upper body, portrait.
| Model | What the Image Shows |
|---|---|
| Base | All features appear. Black tights in the full-body image, a navy blazer with a large red bow in the portrait |
| Nova Anime AM v6.0 | Hair and eyes appear. The tie becomes a thin red ribbon in the full-body image and a red bow in the portrait |
| CottonAnima 2.9B_v1 | Hairstyle and eyes appear, but the scrunchie is pink in the full-body image and purple in the portrait. A red bow at the neck in the portrait |
| AMIXA2.9B 2.9B_V01 | The features appear. The scrunchie is bluish purple. A thin red ribbon at the neck in the full-body image, a red bow in the portrait |
| JANIMA v1.0 2.9B | All features appear. A thin red ribbon at the neck in the portrait |
With the remapped LoRA, all four rendered Kana’s hairstyle as the right character.
Only CottonAnima turned the scrunchie pink or purple, in both the full-body image and the portrait. In the two-character image, it came out blue.
Kei-chan
I generated a full-body standing pose.
For the portraits, I changed the prompt to upper body, portrait.
| Model | What the Image Shows |
|---|---|
| Base | Hair features and the red ribbon tie all appear. Plaid skirt |
| Nova Anime AM v6.0 | The hair appears, but the ribbon at the neck turns navy |
| CottonAnima 2.9B_v1 | The hair appears, but the ribbon at the neck turns navy |
| AMIXA2.9B 2.9B_V01 | The hair features appear. The ribbon at the neck is red in the full-body image and navy in the portrait |
| JANIMA v1.0 2.9B | The hair features appear. The ribbon at the neck is red in the full-body image and navy in the portrait. A navy blazer in the full-body image |
All four also rendered Kei’s hairstyle.
None of them kept the neck ribbon red in both the full-body image and the portrait. AMIXA2.9B and JANIMA had it red in the full-body image and navy in the portrait.
Nova and CottonAnima turned it navy or blue in both images, the same as official Turbo v1.0 and v1.1 in the October 21 article. The prompt’s blue ribbon is meant for the hair ribbon, but it may be pulling the neck ribbon’s color too, but with a single seed-42 image per model, I can’t tell that apart from random variation.
Prompt (Kana & Kei)
Kana full body
masterpiece, best quality, safe, 1girl, solo, kanachan, brown hair, side ponytail, ahoge, brown eyes, school uniform, full body, standing, looking at viewer, white background, simple backgroundKei full body
masterpiece, best quality, safe, 1girl, solo, keichan, blonde hair, long hair, blunt bangs, blue ribbon, blue eyes, school uniform, full body, standing, looking at viewer, white background, simple backgroundFor the portraits, I replaced full body, standing with upper body, portrait and added full body, knees, feet, lower body to the negative prompt.
Negative
worst quality, low quality, blurry, lowres, jpeg artifacts, bad hands, bad anatomy, fused fingers, fused legs, fused limbs, extra arms, extra legs, conjoined twins, deformed faceTwo Characters
Instead of a pose with physical contact, I placed them side by side with standing side by side, full body.
I also added English sentences to the prompt describing who stands on which side and each character’s appearance.
| Model | What the Image Shows |
|---|---|
| Base | Kei on the left, Kana on the right, and Kana’s ahoge does not show up on Kei. Kei has a red ribbon, Kana wears black tights |
| Nova Anime AM v6.0 | Same separation and ahoge. Kei has a navy ribbon, Kana wears black tights |
| CottonAnima 2.9B_v1 | Same separation and ahoge. Kei has a navy ribbon, Kana has a blue scrunchie and a red tie |
| AMIXA2.9B 2.9B_V01 | Same separation and ahoge. Kei has a thin red ribbon and a plaid skirt, Kana has a red tie and black tights |
| JANIMA v1.0 2.9B | Same separation and ahoge. Kei has a beige vest and a red bow, Kana has a red tie and black tights |
All four kept two separate characters on the left and right, and Kana’s ahoge did not show up on Kei. AMIXA2.9B and JANIMA also kept Kei’s ribbon red.
In the 2B article, RDBT, one of the 2B derivatives, grew the scene to four people, but none of the four 2.9B models added extra characters.
Prompt
masterpiece, best quality, safe, 2girls, multiple girls, two different girls, kanachan, keichan, standing side by side, full body, looking at viewer, smile, school uniform, keichan on the left has long blonde hair with blunt bangs and a blue ribbon, kanachan on the right has brown hair with a side ponytail and an ahoge, white background, simple backgroundThe negative prompt is the one from Kana and Kei with 1girl, solo added.
Background Density
I seated Kana-chan by the window of a café.
| Model | What the Image Shows |
|---|---|
| Base | Front-facing composition holding a cup in both hands. Two ceiling lamps, curtains, chairs in the back |
| Nova Anime AM v6.0 | Holds a cup to her mouth, with a second cup on the table. Cars outside the window, a bookshelf and a teapot on a shelf, three potted plants |
| CottonAnima 2.9B_v1 | Holds a cup in both hands. A window, two potted plants, a bookshelf, a ceiling lamp, a picture frame, a chalkboard. A signature-like scribble in the lower left |
| AMIXA2.9B 2.9B_V01 | Holds a cup, with a second one on the table. A window, curtains, a potted plant, a bookshelf, a teapot on a shelf |
| JANIMA v1.0 2.9B | Holds a cup to her mouth, with a second one on the table. Four potted plants, a bookshelf, cups on a shelf |
All four filled the background down to bookshelves and potted plants, with about as much detail as Base.
Nova, AMIXA2.9B, and JANIMA drew a second cup on the table besides the one in her hand, so the single word teacup produced two cups.
Prompt
masterpiece, best quality, safe, 1girl, solo, kanachan, brown hair, side ponytail, ahoge, brown eyes, school uniform, sitting at a cafe table by the window, potted plants, bookshelf, teacup, warm afternoon sunlight, detailed background, intricate details, looking at viewerNegative
worst quality, low quality, blurry, lowres, jpeg artifacts, bad hands, bad anatomy, deformed faceImage-to-Image (i2i)
As in the April WAI-Anima article, I ran a source image made with WAI-Illustrious v160 + the Kana-chan LoRA through image-to-image (i2i) at denoise 0.5. Denoise sets how strongly the model redraws the source, from 0 (unchanged) to 1 (ignore the source), so 0.5 is a middle setting.
The source image used the Kana-chan LoRA, but no character LoRA was applied to the Anima models during i2i.
| Model | What the Image Shows |
|---|---|
| Base | Composition, circle, and sash knot unchanged, with a pattern added inside the circle |
| Nova Anime AM v6.0 | Composition kept, with a plant pattern inside the circle |
| CottonAnima 2.9B_v1 | Composition kept, a white hood added. The background turns pale pink |
| AMIXA2.9B 2.9B_V01 | Composition kept, a white hood added. The background turns pale pink |
| JANIMA v1.0 2.9B | Composition kept, a white hood added. The background turns pale pink |
All five models, including Base, kept the composition, the gold circle, and the sash knot from the source image. What changed was the pattern inside the circle, the white hood, and the background color.
Prompt
masterpiece, best quality, safe, 1girl, solo, brown hair, side ponytail, ahoge, brown eyes, white robe, gold embroidery, capelet, gold sash, long sleeves, long dress, standing, looking at viewer, full bodyThe negative prompt is the same as in Default Style.
Generation Speed
These are the seconds from when the generation script submitted a job to when it received the image, on the same M1 Max 64GB.
Test 1 is the first image for each model and includes model loading. The Base numbers come from the October 21 article.
| Model | Steps | Test 1 | Test 2 | Portrait (832×1024) |
|---|---|---|---|---|
| Base | 30 | 262.8 s | 258.7 s | 210.7 s |
| Nova Anime AM v6.0 | 30 | 383.8 s | 367.7 s | 295.4 s |
| CottonAnima 2.9B_v1 | 30 | 387.0 s | 373.0 s | 297.1 s |
| AMIXA2.9B 2.9B_V01 | 30 | 379.5 s | 371.4 s | 297.2 s |
| JANIMA v1.0 2.9B | 30 | 371.5 s | 371.4 s | 297.1 s |
The four 2.9B models took about 1.4 times as long as Base at the same 30 steps, and all four took nearly the same time.
When I measured plain Anima-2.9B in August, it was also about 1.4 times slower at 30 steps, a similar ratio.
That said, the ComfyUI versions also differ, so I can’t say the gap comes from the block count alone.
Which One for What
Assuming you apply a Base-trained character LoRA, here is how these seed-42 outputs split by use case.
| Use Case | Better Fit in These Tests | Basis |
|---|---|---|
| Placing two characters side by side | All four | Kept two characters on the left and right, and Kana’s ahoge did not show up on Kei |
| Getting Kei’s outfit colors closer | AMIXA2.9B, JANIMA | The ribbon tie stayed red in the full-body and two-character images. It turns navy in portraits |
| Matching Kana’s scrunchie color too | The three other than CottonAnima | CottonAnima made the scrunchie pink and purple in the full-body image and portrait. AMIXA2.9B leans bluish purple |
| Keeping anime-style shading without a character | Nova Anime AM v6.0, AMIXA2.9B | Kept anime shading in both Test 1 and Test 2. JANIMA went semi-realistic in Test 2 |
References
- Nova Anime AM (Civitai)
- CottonAnima (Civitai)
- AMIXA2.9B (Civitai)
- JANIMA (Civitai)
- Anima (Hugging Face)
- October 2026 list of the latest Anima derivative checkpoints
- Testing Official Anima Turbo v1.1 and Aesthetic v1.1 on M1 Max to See If Base-Trained Character LoRAs Work
- Anima-2.9B (Hugging Face)
- Do existing character LoRAs work on Anima-2.9B, expanded to 40 layers?
- Comparison of seven Anima derivative 2B versions
Bonus: NSFW Output Test
Images in this section contain sensitive content from test outputs and have been blurred.
Without the character LoRA, I generated images under two conditions: without the nsfw tag and with it (plus straight-on for a front-facing composition).
I removed safe from the positive prefix.
Without the nsfw Tag
With the nsfw Tag (straight-on)
All five models shown, including Base, produced nudity regardless of whether the tag was present.
Prompt
Without tag
masterpiece, best quality, 1girl, solo, long blonde hair, blue eyes, completely nude, naked, bare skin, sitting on edge of spring, feet in water, forest, natural light, serene, full body, looking at viewerWith tag
masterpiece, best quality, 1girl, solo, long blonde hair, blue eyes, completely nude, naked, bare skin, nsfw, straight-on, full body, standing, white backgroundThe negative prompt is the same as in Default Style.