PonyUltimate Introduction
PonyUltimatePro is an all-in-one ComfyUI generation node designed for Pony Diffusion workflows. It brings prompt preparation, style selection, image sizing, and optional face detailing into one place. Instead of assembling these steps separately for every image, you can set them in a single node.
The node automatically adds Pony scoring tags to your prompts, offers ready-made aesthetic styles and SDXL image sizes, and can detect and refine faces after generation. It can also produce a side-by-side image so you can judge whether the face detailing improved the result.
How PonyUltimate Works
Think of PonyUltimatePro as a small production workflow in one node:
- You describe the image. Enter what you want in the positive prompt and what you want to avoid in the negative prompt.
- The node prepares the prompts. It adds
score_9,score_8_up, andscore_7_upto the positive prompt, andscore_5_upandscore_4_upto the negative prompt. You do not need to type these tags yourself. - You choose the look and shape. Select an aesthetic style and an aspect-ratio preset, or choose
Nonefor the style. - The image is generated. The node uses your selected checkpoint and generation settings.
- Faces can be refined. When face detailing is enabled, the node uses face detection to locate faces and FaceDetailer to work on them. You can optionally view the result beside the base image.
For example, you could write a prompt for an anime character, choose the Anime style and Portrait dimensions, then turn on face detailing to see whether it improves the character’s face.
PonyUltimate Features
Automatic Pony Scoring Tags
PonyUltimatePro adds scoring tags to both prompts automatically:
| Prompt | Tags added by the node |
|---|---|
| Positive | score_9, score_8_up, score_7_up |
| Negative | score_5_up, score_4_up |
Write the subject, setting, and other creative details you want; the node handles these tags for you.
Aesthetic Style Presets
The style dropdown offers Cinematic, Photographic, Digital Art, Anime, and Vintage modifiers. These give you a quick way to try a different aesthetic without rewriting your whole prompt. Select None when you do not want a style preset applied.
A preset guides the image’s appearance, but your checkpoint, prompts, and generation settings also influence the result. Try the same prompt with two styles to see which better suits your idea.
SDXL Size Presets
The aspect_ratio dropdown provides these image dimensions:
| Preset | Dimensions | Useful for |
|---|---|---|
| Square (1:1) | 1024 × 1024 | Balanced, square compositions |
| Portrait | 832 × 1216 | Taller character-focused images |
| Landscape | 1216 × 832 | Wider scenes |
| Vertical (9:16) | 768 × 1344 | Tall, narrow compositions |
| Wide (16:9) | 1344 × 768 | Wide, cinematic compositions |
Changing the size changes the space available for your subject and background. If a composition feels cramped, try a wider preset; if you want more room for a standing character, try a taller one.
Integrated Face Detailing
Set face_detailer to enable or disable automated face refinement. When enabled, the node uses FaceDetailer with an Ultralytics YOLO face detector to find faces in the generated image.
The face_denoise setting controls how strongly the detailing step alters a face. If you want to preserve the original look more closely, try a gentler setting; if the face needs more noticeable repair, try a stronger one. Check the result visually, since a stronger change is not always a better one.
Before-and-After Comparison
Set show_comparison to Yes to output a side-by-side view of the base image and the face-detailed image. This makes it easier to decide whether to keep face detailing enabled or adjust face_denoise. When comparison is not enabled, the IMAGE output is the final generated image.
Generation Controls and Outputs
You can select a checkpoint with ckpt_name and enter your main prompts in positive and negative. The node also exposes standard generation controls: seed, steps, cfg, sampler_name, and scheduler. These let you adjust the generation process while keeping the same combined workflow.
PonyUltimatePro provides three outputs:
| Output | What it contains |
|---|---|
IMAGE | The final image, or the side-by-side view when comparison is enabled |
MODEL | The loaded checkpoint model, available to other nodes |
VAE | The loaded VAE, available to other nodes |
PonyUltimate Models
PonyUltimatePro does not specify a fixed list of included image-generation checkpoints. You choose a checkpoint through ckpt_name; the original node configuration defaults to a path in a Pony model folder. Your chosen checkpoint is the model used for generation and is also available through the MODEL output.
Face detailing uses a separate model for a different job: face_yolov8m.pt detects where faces are in the image. It is not the checkpoint that creates the image. The node relies on the FaceDetailer and UltralyticsDetectorProvider classes from ComfyUI-Impact-Pack for this face-detection and detailing workflow.
Troubleshooting PonyUltimate
The node reports a missing FaceDetailer or UltralyticsDetectorProvider
- Check whether ComfyUI-Impact-Pack is available in your ComfyUI environment.
- Confirm that its FaceDetailer and UltralyticsDetectorProvider nodes or classes are available.
- Try loading your workflow again after resolving the missing dependency.
PonyUltimatePro calls these external classes, so the face-detailing workflow cannot function without them.
Face detection reports a missing model
Check that the required face_yolov8m.pt bounding-box model is available to the detector. This model is used to locate faces, not to generate the image.
The checkpoint cannot be selected or loaded
Check the value of ckpt_name and select a checkpoint that is available in your ComfyUI environment. The default points to a Pony model folder path, which may not match your own checkpoint collection.
Face detailing changes the face too much
- Turn on
show_comparisonso you can inspect the base and detailed images together. - Try a gentler
face_denoisesetting. - Disable
face_detailerif you prefer the original face.
Frequently Asked Questions
Do I need to add the Pony score tags to my prompts?
No. The node adds the listed positive and negative scoring tags automatically.
Can I use my own prompt with a style preset?
Yes. Enter your prompt as usual, then choose a style from the dropdown. Select None when you do not want a preset.
Does the comparison setting change the image output?
Yes. With show_comparison enabled, the IMAGE output is a side-by-side view rather than only the final image.
Are the checkpoint and the face detector the same model?
No. The checkpoint selected with ckpt_name generates the image. face_yolov8m.pt helps the face-detailing process locate faces.
