Published October 21, 2026by DXMPie

Anima Made Easy: The Full Guide

359 views5 reactions0 comments on CivitAI8 collected

loradetailrefinerwanimage to videotext to videoguidetool guideanimaanimeworkflowcomfyuiregional promptingadetailerupscaleinpaintimg2img

Anima Made Easy Brain Edition: The Full Guide

A complete walkthrough of the workflow: what every group does, how to set it up, and how to fix the usual problems.

This guide covers the expanded version with draft-then-hi-res, regional prompting, batch tools and Wan 2.2 and MiniMax H3 video.

Contents

Overview
Requirements
Quick start
The layout and jump keys
Group by group
Draft then hi-res, and the size chart
Prompting
Regional prompting in detail
Video
Posting your pictures to Civitai
Troubleshooting
The Brain Edition: batch tools and overnight runs

DOWNLOAD BELOW!

https://civitai.red/models/3013250/anima-made-easy-complete-comfyui-workflow-with-wan-video

The download has TWO workflows. Same install, use whichever you like.

Guided (Anima Made Easy): notes next to every box, jump keys, everything explained. Best for learning it.
Brain Edition (new in v3.0): no notes in the way. One central prompt with every control you need around it. Built for running a lot of pictures. Its full how-to is section 12.

1. Overview

The workflow is one big canvas split into coloured groups. Each group is one job, and each one can be switched on or off. Off groups are skipped and cost nothing.

A1 (key 2): Models, LoRAs, prompts. Always on. In the Brain Edition this is the Brain, with the global prompt in the middle and every control around it.
A2 (key 3): Text to image drafts. Needs A1.
B (key 4): Hi-res pass. Needs A1 and A2 (cached).
C (key 5): Adetailer for face, eyes and hands. Needs A1 and a picture.
C0 (no key): Load Image for C. Only when B is off.
D (key 6): Tiled upscale. Needs A1 and a picture.
E (key 7): Inpainting. Needs A1, a picture and a mask.
F (key 8): Simple model upscale. Needs a picture.
G (key 9): Image to image. Needs A1 and a picture.
H1 (key 0): Regional draft. Needs A1.
H2 (no key): Regional hi-res. Needs H1 (cached).
I: Batch Upgrade. Hi-res a whole folder of drafts (section 13).
J: Batch Final Upscale. Enlarges a whole batch folder by shape (section 13).
K: Prompt file loader. Runs a list of prompts overnight (section 13).
V1: Wan text to video. Needs video models.
V2: Wan image to video. Needs video models and a picture.
V3: Wan first and last frame to video. Needs video models and two pictures.
V2x: Wan chain. Continues a clip from its last frame for a longer video.
V4: MiniMax H3 image to video, with sound. Needs the H3 model files.

The key idea: A1 stays on, you switch the other group you want on, and you press Run.

2. Requirements

Custom nodes

Install with ComfyUI-Manager (search by name):

ComfyUI-Impact-Pack (ltdrdata): wildcards, FaceDetailer.
ComfyUI-Impact-Subpack (ltdrdata): Ultralytics detectors.
rgthree-comfy (rgthree): group switches, jump keys, Any Switch.
ComfyUI-Lora-Manager (willmiao): the LoRA loader in A1.
ComfyUI_UltimateSDUpscale (ssitu): group D.
ComfyUI-ppm (pamparamm): Attention Couple, regional prompting.
ComfyUI-GGUF (city96): the Wan video groups only.
anima_batch_loader (included in the download, not in the Manager): the prompt file loader, the batch groups and the K key. Copy the whole folder into ComfyUI/custom_nodes/ and restart.

Update ComfyUI itself first. The Wan and MiniMax H3 video nodes are built in to recent versions (H3 needs ComfyUI 0.38 or newer).

Models checklist

Anima model goes in models/diffusion_models: https://huggingface.co/circlestone-labs/Anima or https://civitai.com/models/2458426/anima-official
qwen_3_06b_base.safetensors goes in models/text_encoders (Anima repo).
qwen_image_vae.safetensors goes in models/vae (Anima repo).
face_yolov9c.pt goes in models/ultralytics/bbox: https://huggingface.co/Bingsu/adetailer/blob/main/face_yolov9c.pt
hand_yolov9c.pt goes in models/ultralytics/bbox: https://huggingface.co/Bingsu/adetailer/blob/main/hand_yolov9c.pt
Eyeful_v2-Individual.pt goes in models/ultralytics/bbox: https://civitai.com/models/178518/eyeful-or-robust-eye-detection-for-adetailer-comfyui
4x_foolhardy_Remacri.pth goes in models/upscale_models. Any 2x or 4x upscaler works.

Wan video (optional):

Wan2.2-I2V-A14B-HighNoise-Q4_K_M.gguf and ...LowNoise-Q4_K_M.gguf (QuantStack/Wan2.2-I2V-A14B-GGUF) go in models/diffusion_models.
Wan2.2-T2V-A14B-HighNoise-Q4_K_M.gguf and ...LowNoise-Q4_K_M.gguf (QuantStack/Wan2.2-T2V-A14B-GGUF) go in models/diffusion_models.
The 4-step LoRAs, I2V pair and T2V pair (lightx2v/Wan2.2-Distill-Loras), go in models/loras.
wan_2.1_vae.safetensors goes in models/vae.
umt5_xxl_fp8_e4m3fn_scaled.safetensors goes in models/text_encoders.

Download single files. The original Wan-AI repositories are 34 GB and 68 GB of shards that ComfyUI cannot use directly.

MiniMax H3 video (optional): see section 13 for the files and the setup.

If a model box says Value not in list, click it and pick the file you have. That is normal the first time.

3. Quick start

Install the custom nodes above, then copy the anima_batch_loader folder into ComfyUI/custom_nodes/ and restart ComfyUI.
Copy everything in the download's input folder into ComfyUI/input/. It has the regional masks, an example picture and an example prompts.txt.
Drag a workflow onto ComfyUI (guided or Brain Edition).
Pick your Anima model, text encoder and VAE in A1.
Press 2. Make sure A1 and A2 are on and everything else is off.
Type a prompt in the positive box.
Press Run. Drafts appear in Anima/draft/ with the seed in the file name.
Keeper? Guided version: drag its PNG onto ComfyUI, set the DraftSampler's control after generate to fixed, switch B on, press Run. Only B runs. Brain Edition: use the hi-res button for that shape.

4. The layout and jump keys

The canvas is a grid. Press a number key (the number row, not the numpad) to fly to a group. The group lands at the left of your screen with its on/off switch beside it.

1: Home, the control panel.
2: A1, the Brain (Models, Prompts and Sizes).
3: I, Batch Upgrade (folder).
4: H1, Regional Prompting.
5: G, Image To Image.
6: V2, Image to Video.
7: C, Adetailer and its Load Image.
8: D, Upscale.
9: E, Inpainting.
0: F, Simple Upscale.
minus: H2, Regional Hi-res.
equals: V1, Text to Video.

Text to Image and Hi-res have no jump key. Their on/off boxes and sizes sit under the prompt, so key 2 covers them. You can also click the bookmark tab at the top to open the full shortcut list instead of pressing keys.

Two more keys work on a picture instead of a group: hover a saved picture and press K to send it to keepers.

If a key does nothing, click an empty spot on the canvas first. For a mouse-free overview, turn on ComfyUI's minimap (bottom right) and click any spot.

Switches

Control panel (Home): lists every group with an on/off toggle.
Beside each group: a small switch for that group.
Next to the draft (A2): a quick panel for B and C, so you can turn the hi-res pass and adetailer on without jumping.
Next to regional prompting (H1): a copy of the all-groups panel, so you can turn the others off from there.
Video Controls (Brain Edition): a box beside the Brain with the video picture, the V switches and the H3 prompt, so you never have to go down to the video groups.

Passing a picture along

Right-click a result in any Save Image box, choose Copy (Clipboard), click a Load Image box in another group and press Ctrl+V. That is the fastest way to move a picture between groups.

5. Group by group

A1. Models and Prompts (always on)

Holds the Anima model, the Qwen text encoder, the Qwen image VAE, the LoRA loader, and the positive and negative prompt boxes. Every other group uses these, so keep A1 on.

Type <lora:name:0.5> in the LoRA box, or use the picker.
The prompt boxes are wildcard boxes. Write __name__ to pull a random line from a wildcard file.

A2. Text to Image

Makes the draft. Default size 768x768, 35 steps, CFG 3, er_sde, simple scheduler. Drafts are small on purpose: they are fast, and the hires pass turns the keepers into the final.

Keep batch size at 1 and queue several runs. Each picture then gets its own seed.

B. Hires Pass

Enlarges the draft and redraws it with more detail. Default final size 1248x1248, denoise 0.55.

Denoise Result 0.45: Barely changes the draft; safest 0.55 adds detail, small drift (start here) 0.65 and up The picture starts to change

Details are in the next section.

C. Adetailer

Three detailers run one after another: face, eyes, hands.

Part Default denoise Face 0.26 Eyes 0.24 Hands 0.40

If hands come out worse, lower the hands denoise.

Where does the picture come from? In this order:

B's final picture, if B is on.
The picture in C0, if C0 is on.
The A2 draft.

So you can run A2 + C alone to adetail a fresh draft, A2 + B + C for the full chain, or C0 + C to adetail a saved picture.

To adetail a saved picture: switch C0 on and load it, switch B and A2 off, switch C on.

D. Upscale Advanced

Tiled Ultimate SD Upscale. Default: 2x, 18 steps, CFG 6, dpmpp_2m_sde karras, denoise 0.14, tile 512. It adds detail and keeps the picture. If you see seams, raise seam_fix_denoise. The upscale model's name usually says 2x or 4x, but upscale_by in the node is the size you get.

E. Inpainting

Load the picture in the Load Image box.
Right-click it and choose Open in MaskEditor. Paint what you want changed and save.
In the inpaint prompt write only what goes inside the mask, for example closed mouth, smiling.
Run. Default denoise 0.6. Use 0.75 for a bigger change and 0.4 for a small fix.

Differential Diffusion softens the mask edge. The result is pasted back onto the original, so everything outside the mask keeps its quality.

F. Simple Upscale

Model upscale only. No sampler, no new detail. A 4x model makes the picture four times wider. The Scale Down node at 0.5 gives a clean 2x. Set it to 1.0 for the full 4x.

G. Image To Image

Load a picture, write a prompt, run. Denoise decides how much changes:

Denoise Result 0.3 Small touch-ups 0.5 Some change (start here) 0.7 A lot of change

Keep the picture around 1 megapixel (1024x1024). Much bigger can stretch or double things.

H1 and H2. Regional prompting

Its own engine with its own prompts, sampler and save. See section 8.

V1 and V2. Video

See section 9.

6. Draft then hires, and the size chart

The draft and the hires pass use the same seed. Draft runs use randomize. When you pick a keeper, drag its PNG onto ComfyUI and set the DraftSampler to fixed. The hires pass then uses the same seed, so the final is the same picture, only sharper.

Rules:

The draft and the hires pass need the same steps (35). Make drafts faster by using a smaller size, not fewer steps.
Always pick a draft size and final size from the same row, so the proportions match.
If the final looks different from the draft, the sizes probably did not match a pair.

Size chart

Shape Final Draft Scale Square 1248x1248 768x768 1.625 Wide 1792x768 1120x480 1.6 Wide 1536x1024 960x640 1.6 Wide 2304x768 1440x480 1.6 Wide 1664x960 832x480 2.0 Portrait 1152x1536 720x960 1.6 Portrait 1024x1536 640x960 1.6 Portrait 768x1792 480x1120 1.6 Portrait 768x2304 480x1440 1.6 Portrait 960x1664 480x832 2.0

Want your own pair? Pick a draft size, multiply both numbers by 1.6 or 2, and keep the results as multiples of 16.

7. Prompting

Anima likes this tag order:

[quality / meta / year / safety tags] [1girl / 1boy / 1other] [character] [series] [@artist] [general tags]

Example:

masterpiece, best quality, safe, 1girl, solo, long hair, smile, looking at viewer, outdoors, sunny day

Settings that work: 30 to 50 steps, CFG 3 to 5, about 1 megapixel, sampler er_sde, euler_a or dpmpp_2m_sde.

Tips:

An @artist tag and a LoRA or two make a big difference. A bare prompt looks generic.
Describe every garment, including the bottoms.
Simple beats complicated. Around 45 to 70 tags is plenty.

8. Regional prompting in detail

Regional prompting lets you describe several people in one picture, each in their own area, without their traits bleeding into each other.

How it works

You load a mask picture. Each colour in it is one region:

Colour Position in the 3 and 4 region masks Red Left Green Second from left Blue Third from left Yellow Fourth, far right

Each colour has its own prompt box, in the same order, plus a BASE prompt at the top.

BASE prompt: style and the shared scene only, for example masterpiece, best quality, outdoors, beach, sunny day. Never describe a character here, because it applies to the whole picture and bleeds into every region.
RED, GREEN, BLUE, YELLOW: one person each, with outfit and pose.
Models, LoRAs, negative prompt and VAE come from A1, so keep A1 on.
Switch A2 off while you use H1, or it makes a normal draft too.

The ready-made masks

Three mask pictures come with the workflow, all 3584x1536:

File Characters Layout Anima_Mask_2_Regions_3584x1536.png 2 Red left, green right Anima_Mask_3_Regions_3584x1536.png 3 Red, green, blue Anima_Mask_4_Regions_3584x1536.png 4 Red, green, blue, yellow

Put them in ComfyUI's input folder, then pick one in the Load Mask Image box. Using the 2-region mask? Leave the BLUE and YELLOW prompts alone, they do nothing.

The black gaps between regions are not painted, so only the base prompt applies there. That keeps the people apart.

Make your own mask

Any picture works. Paint solid colours with no soft edges: pure red (255,0,0), green (0,255,0), blue (0,0,255) and yellow (255,255,0), on black. Yellow is built from red and green painted together, so paint it as yellow, not as a mix. Any size works: the workflow reads your picture's size and matches everything to it.

Sizes and hires

H1 makes the draft (default 1120x480) and H2 makes the final (default 1792x768). Use H2 the same way as B: drag the keeper PNG onto ComfyUI, set the RegionalSampler to fixed, keep H1 on, switch H2 on, press Run.

Tips

Put the biggest or most important person in the roomiest region.
Keep poses simple. Two people touching each other is hard for any model.
If a region ignores its prompt, make the mask region bigger.

9. Video (Wan 2.2 14B)

Both video groups use two models, a high noise model that draws the motion and a low noise model that adds detail. Each has a 4-step LoRA, so a clip takes only 4 steps (cfg 1, euler).

V2. Image to Video

Load a picture. It becomes the first frame.
Describe only what moves, for example she blinks and smiles, her hair moving in the breeze.
Switch A2, B and C off and V2 on. Run.

Use the I2V models and I2V LoRAs.

V1. Text to Video

Describe the whole scene and the motion in plain sentences. Anima does not run here, so no tags. Use the T2V models and T2V LoRAs.

Length

Clip length = frames divided by fps.

Frames Seconds at 16 fps: 49 about 3, 81 about 5 97 about 6 113 about 7

Frames must be 4 times something plus 1. Do not raise the fps to get a longer clip. That only shortens the clip. Raise the frames instead, and stay under about 121. Width and height must be multiples of 16, and 832x480 is a good start.

For longer videos, save the last frame of a clip and use it as the start picture of the next one.

10. Posting your pictures to Civitai

Civitai fills in the Resources box (model and LoRAs) from the metadata inside the picture. ComfyUI's normal Save Image node does not write that. The Save Image (LoraManager) node does, and it comes with the Lora Manager pack you already installed.

In this workflow, every image group saves with that node. It also keeps the whole workflow inside the PNG, so dragging a picture back onto ComfyUI still restores everything.

For the Resources box to fill in:

Your Anima model and your LoRAs must be in Lora Manager's library, and Lora Manager must know them from Civitai (open the Lora Manager page and let it fetch the Civitai info for each one).
Generate, then upload the PNG to Civitai.

What to fill in on the image page:

Field What to put Prompt Leave it on and show the prompt. Civitai reads it from the picture. Resources Filled in automatically. If something is missing, press + Resource and add it. Tools ComfyUI Techniques Add what you used, for example txt2img, hires fix, upscale, inpainting, img2img

Pointing people to the workflow: publish your images on Civitai at the Workflow page. Scroll to where the images are and use the 'add post' button. This will tie it to the workflow and show all your pretty pictures off on the workflow page! Civitai only auto-detects resources it can match by hash, and a workflow file has no hash in a picture, so this part is manual.

Final images are saved with the same name pattern as before, but the seed in the file name comes from the sampler that made that picture. For finals that is the hires pass, so keep your keeper's draft seed in the draft file name.

11. Troubleshooting

Problem Cause and fix Required input missing or Missing connection A group the current one needs is off. Keep A1 on.

Adetailer: NoneType has no len The picture source is empty. Turn B on, or turn C0 on and load a picture, or turn A2 on.

Value not in list A file name differs from yours. Click the box and pick your file.

stack expects each tensor to be equal size The mask sizes differ. The workflow sizes them for you, so this means the mask loader or base mask was changed. Reload the workflow.

Final looks different from the draft Seed not fixed, steps differ, or the sizes are not a matching pair.

Video is noisy or crinkled texture The model and LoRA types do not match. Use T2V models with T2V LoRAs, and I2V with I2V.

Video is only 2 seconds The fps is too high. Use 16 fps and raise the frames.

GGUF node missing (red node) Install ComfyUI-GGUF. It is only needed for the video groups.

Resources box is empty on Civitai Make sure Lora Manager has the Civitai info for your model and LoRAs. A key does nothing Click an empty spot on the canvas, then press the key.

Seams in the tiled upscale Raise seam_fix_denoise or the tile size in D.

Hands got worse after C Lower the hands denoise in C.

The floating toolbar over a node is annoying ComfyUI Settings, search "selection toolbox", turn it off.

12. The Brain Edition: batch tools and overnight runs

The Brain Edition adds a few tools for running a lot of pictures without babysitting. Everything in this section is optional. The normal draft then hires flow in sections 3 to 6 works exactly as before.

What you need

The batch tools use one custom node folder, anima_batch_loader. Copy the whole folder into ComfyUI/custom_nodes/ and restart ComfyUI once. It adds:

the Anima Batch Folder Loader (reads a folder of pictures one at a time)
the Anima Prompt File Loader (reads prompts from a text file, one per run)
the Anima Save nodes (keep the prompt and LoRAs inside the picture)
two keyboard shortcuts: K (send to keepers)

If you only use the normal groups you do not need it, but the workflow will show a missing node warning without it.

Keepers: the K key

Hover any saved picture in ComfyUI and press K, or right-click a Save node and choose "Send to keepers". The picture is copied to ComfyUI/output/Anima/keepers. You can change the folder name in ComfyUI Settings, in the Anima section.

Prompt file: run a whole list overnight

Group K sits to the left of the brain.

Put a text file called prompts.txt in ComfyUI/input.
Separate prompts with a line of five or more = signs. A first line starting with ## is a title and is not sent to the model. If the file has no divider lines, every line is its own prompt. Blank lines are ignored.
In the PROMPT FILE loader turn use_file ON and set the file name.
repeats is how many pictures you want per prompt.
Set PROMPT FILE INDEX to 0. Set Batch count (next to the Run button) to prompts x repeats.
Press Run. When the list is finished, the run stops with a "finished" message. That is normal.

You can add prompts to the bottom of the file while it is running. Turn use_file OFF to go back to typing prompts yourself.

Example file:

## Garden girl
(masterpiece, best quality, very aesthetic:1.1), 1girl, solo, ...
=====
## Night rooftop
(masterpiece, best quality, very aesthetic:1.1), 1girl, solo, ...

Batch Upgrade (group I)

Hires a whole folder of keepers at once.

Fill Anima/keepers with the drafts you liked (K key).
Turn group I on with the BATCH buttons in the brain. Everything else stays off.
Set BATCH HIRES INDEX to 0 and Batch count to the number of pictures. Press Run.

Each picture is read with its own prompt and sized for its shape (phone, wallpaper or square) using the same size boxes as the normal hires. Results go to Anima/batch_final. Empty the keepers folder when you are done.

Batch Final Upscale (group J)

Reads Anima/batch_final, enlarges each picture and sizes it by shape with the PHONE, WALLPAPER and SQUARE boxes (1.25 means 125%). Results go to Anima/final_upscaled with the prompt and resources kept, so Civitai still shows your model and LoRAs. Set UPSCALE INDEX to 0 and Batch count to the number of pictures first.

The index boxes

The batch boxes count up by one every run, so they must be set back to 0 before each new batch. The quick way: open the command palette and run Anima: set the batch index boxes to 0. If a batch seems to repeat or skip pictures, this is almost always the cause. Also check you do not have the workflow open in two tabs with different index values.

Seeds

Drafts take their seed from the DRAFT SEED box. randomize gives a new picture each run, fixed locks it. If your batch keeps producing the same number, set it to randomize and restart ComfyUI once. The hires samplers keep their own fixed seeds on purpose, so a hires always matches its draft.

Troubleshooting the batch tools

"finished" error: the file or folder ran out. Batch count was higher than the number of items. Normal, nothing is wrong.
File not found: the prompt file must be inside ComfyUI/input, or you must type its full path.
Same picture twice: index not reset to 0, or the seed is on fixed.
Embedding or LoRA text in a prompt crashes the encoder: Anima does not read SDXL embeddings or <lora:...> tags typed in the prompt. Remove them from the text and add LoRAs in the LoRAs box.

v1 First release: draft and hires, adetailer, upscale, inpaint, image to image, regional prompting for 2 to 4 characters, Wan 2.2 14B text to video and image to video.

## UPDATE v2.0: portrait, batch upgrade, and Send to keepers

This is a big one. Everything is in the v2.0 download.

What's new

- Portrait pair. A2 and B now come in Wide, Square and Portrait (default 896x1152 up to 1344x1728).

- Sizes in one place. Under the prompt, there is a chart of ready-made size pairs and a row of boxes per shape. Type the draft and final size you want, and every group follows.

- One small box per shape, next to the prompt. Draft on top, hires below. Adetailer has its own box, so hires no longer switches it on by accident.

- Batch Upgrade (group I). Put your keeper drafts in a folder, set Batch count to the number of pictures, press Run once. Each picture is hired with its own saved prompt, so every character keeps its details. It skips re-making the draft, so it runs about 25 to 30 seconds per picture instead of 45 to 50.

- Send to keepers. Hover a draft (in the feed or on a Save node) and press K, or right-click a Save node and choose Send to keepers. It copies the picture into Anima/keepers for the batch. Change the key under Settings, Keybinding, and search "Anima".

- Jump keys. Key 3 now goes to the sizes, key 4 to Batch Upgrade. You can also click the bookmark tab at the top to see every jump point.

- Regional prompt masks are in the download again (five of them), with the install note.

Install

1. Copy the whole anima_batch_loader folder into ComfyUI/custom_nodes/, restart ComfyUI, and refresh the browser tab. It is only needed for Batch Upgrade and Send to keepers. The rest of the workflow runs without it.

2. Open the v2.0 workflow.

3. For regional prompting, copy the default mask (Anima_Mask_3_Regions_1792x768.png) into ComfyUI/input/.

Good to know

- Batch Upgrade only reads PNGs made by this workflow, because each prompt is read from the picture.

- Set the index box to 0 before every batch and empty the keepers folder afterward.

- In Nodes 2.0, the yes/no state on the switch boxes does not redraw (a limitation of the rgthree switch nodes). They still work. Send to keepers works in both views, but the right-click item may only appear in the regular view. Hover and press K works in either.