ADetailer Single Character Focus - How to Target Specific Faces in Multi-Character Images 2025

You've generated an image with multiple characters, but only one needs face enhancement. ADetailer keeps fixing all faces with the same settings, or worse, applying the wrong characteristics to the wrong people. This guide shows you how to target specific faces for individual treatment.

Quick Answer: In Automatic1111's ADetailer, use the [SEP] keyword to apply different prompts to different faces in order of appearance. In ComfyUI's FaceDetailer (Impact Pack), adjust bbox threshold to target specific faces, or use separate FaceDetailer passes with masks to isolate individual characters.

What Is the Challenge With Multi-Character Faces?

When ADetailer or FaceDetailer runs on images with multiple people, it processes all detected faces with the same settings. This creates problems when characters need different treatments.

Common Issues:

All faces get the same enhancement prompt, causing trait bleeding between characters. Character-specific features (eye color, expressions) apply to wrong people. Quality varies across faces with single settings.

Why This Happens:

Face detailers work by detecting faces and applying enhancement in a loop. Without specific targeting, each face receives identical processing.

How Do You Use [SEP] in ADetailer?

Automatic1111's ADetailer extension supports the [SEP] keyword for applying different prompts to different faces.

Basic Syntax:

In the ADetailer prompt field, separate prompts with [SEP]:

"green eyes, blonde hair [SEP] blue eyes, dark hair [SEP] brown eyes, red hair"

How It Works:

ADetailer processes faces in order of detection (typically left-to-right, larger faces first). First detected face gets first prompt. Second detected face gets second prompt. Additional faces continue the pattern.

Example:

Image has three people: woman on left, man in center, child on right.

Prompt: "beautiful woman, green eyes [SEP] handsome man, beard [SEP] cute child, freckles"

Each face receives its specific description.

Limitations:

Face detection order isn't always predictable. Very similar faces may swap order between generations. Background faces may be detected unexpectedly.

How Do You Target Faces in ComfyUI FaceDetailer?

ComfyUI's FaceDetailer from Impact Pack requires different approaches since it doesn't have [SEP] syntax.

Threshold Adjustment:

The bbox threshold controls which detected faces are processed.

Threshold	Effect
High (0.7+)	Only clear, prominent faces processed
Medium (0.5)	Most visible faces processed
Low (0.3)	Even small/unclear faces processed

Targeting Prominent Faces:

Increase bbox threshold to process only the most clearly defined face. This typically selects the largest or most front-facing face.

Including Background Faces:

Lower threshold to catch smaller faces in crowds or background. Be aware this may include unintended detections.

Sequential Processing:

Run multiple FaceDetailer passes with different settings. First pass targets main character. Second pass targets secondary characters with adjusted thresholds.

How Do Mask-Based Approaches Work?

Masks provide the most precise face targeting.

Creating Character Masks:

Generate masks that isolate each character's face region. Use segmentation, manual drawing, or separate detection passes.

Applying Masks to FaceDetailer:

Connect character-specific masks to FaceDetailer inputs. Each masked FaceDetailer pass only processes the specified region.

Workflow Example:

Use SAM or similar to create masks for each character. Create parallel FaceDetailer branches, one per character. Each branch has its own prompt and settings. Combine results.

Benefits:

Complete control over which face gets which treatment. No reliance on detection order. Works regardless of face positions or sizes.

Drawbacks:

More complex workflow setup. Requires mask creation step. More processing time.

For a simpler workflow, first verify that the chosen tool exposes the face selection or mask control your image needs. A general enhancement feature may not provide those controls.

What Affects Face Detection Order?

Do not infer detection order from a character's identity. The detector, sorting settings and image composition can change which detected region receives each prompt. The ADetailer documentation describes its detection, masking and inpainting options; check the version and extension you are actually using.

Preview the detected regions before assigning prompts. If your workflow permits selecting or drawing a mask, isolate the intended face, inspect mask edges and verify the final image. A mask controls the edit region but does not guarantee identity preservation or a correct result. Moving characters apart can make selection easier; it is not a reliable identity-matching rule.

What Are Common Problems and Solutions?

Users encounter predictable issues with multi-face targeting.

Problem: Wrong Face Gets Prompt

The prompt meant for character A applies to character B.

Solutions:

• Verify detection order by testing
• Select and inspect a mask for the intended face
• Swap prompt order to match actual detection
• Adjust threshold to exclude the wrong face

Problem: Some Faces Not Detected

Character faces are skipped entirely.

Solutions:

• Lower bbox threshold
• Ensure faces are clearly visible
• Check for occlusion or extreme angles
• Use different face detector model

Problem: Background Faces Processed

Unimportant background faces receive enhancement.

Solutions:

• Increase bbox threshold
• Use masks to exclude background
• Use maximum faces limit
• Consider inpainting to remove unwanted faces

Problem: Inconsistent Results Across Generations

Face targeting varies between generations.

Solutions:

• Use masks for consistency
• Fix seeds for more predictable layouts
• Increase detection threshold for stability

How Do You Create Effective Per-Character Prompts?

Prompt construction affects targeting success.

Character-Specific Details:

Include unique identifiers for each character. "Character with glasses [SEP] character with beard" works better than generic prompts.

Quality Terms:

Include quality terms in each prompt section. "detailed eyes, high quality, beautiful woman [SEP] detailed eyes, high quality, handsome man"

Avoiding Conflicts:

Don't use terms that could apply to wrong characters. Keep descriptions mutually exclusive where possible.

Prompt Order Testing:

Generate test images to verify which face receives which prompt. Adjust order based on actual detection behavior.

Frequently Asked Questions

Does [SEP] work in ComfyUI FaceDetailer?

No, [SEP] is specific to Automatic1111's ADetailer. ComfyUI requires threshold adjustment or mask-based approaches.

How many faces can [SEP] handle?

No strict limit, but more faces means more potential for order confusion. Masks become more practical with many faces.

Can I target faces by position?

Not directly. Detection order correlates with position but isn't guaranteed. Masks provide position-based targeting.

What if faces are very similar?

Very similar faces challenge detection ordering. Masks or clear visual differentiators (glasses, hair color) help.

Does this work with anime/stylized faces?

Yes, though detection may be less reliable. Use detectors designed for anime if available.

Can I skip certain faces entirely?

Use higher thresholds to exclude less prominent faces, or use masks to process only specific faces.

How do I handle overlapping faces?

Overlapping faces challenge all approaches. Consider generating with better separation or masking.

What's the best approach for consistent characters?

Masks provide most consistent targeting. For simpler needs, threshold adjustment with consistent positioning works.

Conclusion

Targeting specific faces in multi-character images requires understanding how face detection works and applying appropriate techniques. ADetailer's [SEP] keyword handles simple ordered cases, while ComfyUI requires threshold adjustment or mask-based workflows.

Key Approaches:

Use [SEP] in A1111 for ordered face prompts. Adjust FaceDetailer bbox threshold for selective processing. Create and inspect masks for specific faces, then check that other faces remain acceptable.

Best Practices:

Test detection order before relying on it. Use masks for complex or critical applications. Include character-specific identifiers in prompts. Verify results match intentions.

Getting Started:

Start with threshold adjustment for simple cases. Learn mask creation for precision needs. Document what works for your specific workflows.

Before choosing a hosted tool, check whether it supports selecting the intended face or supplying a mask. General face enhancement is not evidence of individual character targeting.

Mastering face targeting transforms multi-character generation from frustrating to controllable. The right technique depends on your specific needs and workflow complexity tolerance.