ComfyUI Prompt Interview

Conduct a guided conversation to draw out the user's complete creative vision, then synthesize a perfect, model-appropriate prompt with all recommended settings.

When to Invoke This Skill

• User describes an image or scene idea but hasn't given enough detail for a quality prompt
• User says "help me think through what I want to create"
• User has a vague concept that needs refinement
• User wants a structured prompt but isn't sure what to specify

The Interview Philosophy

Ask, don't interrogate. This is a conversation, not a form. Ask one or two questions at a time. Listen to what the user gives you and follow up on what's missing. Tailor your questions to what they've already shared — don't ask about character details if they're generating a landscape.

Fewer questions = better. Aim for 4-7 exchanges maximum. Ask the most impactful questions first. Stop asking when you have enough to generate an excellent prompt.

Don't ask for what you can infer. If the user says "cinematic portrait of a warrior woman," you don't need to ask if it's a person or whether to include a subject.

---

Interview Flow

Step 1: Open with the Big Picture

If the user hasn't told you what they want to create, start here:

"What do you want to create? Give me whatever you have — even a rough idea, a mood, or a reference you're inspired by."

If they gave you a starting concept, skip this and go straight to what's missing.

---

Step 2: Branch by Creation Type

Based on their answer, determine what kind of generation this is:

Type	Key Questions to Ask
Portrait / Character	Identity method? Existing character? Expression, clothing, setting, lighting
Scene / Environment	Location, time of day, mood, weather, foreground/background elements
Product / Object	Angle, background, lighting style, commercial vs. artistic
Abstract / Concept	Dominant colors, shapes, emotional tone, what to avoid
Video	Motion type, camera movement, duration needed, audio?

---

Step 3: Ask the High-Impact Questions

Ask only what's missing. Use natural conversational language, not a bullet list.

For character/portrait content — ask in order of impact:

1. Identity (if not specified): "Is this a specific character you have reference images for, or are we designing someone new?"
2. Expression & mood: "What's the emotion or energy — fierce, serene, playful, haunted?"
3. Setting: "Where are they, and when? (Time of day, location, interior/exterior)"
4. Lighting: "Any specific lighting in mind? (Golden hour, dramatic side light, soft studio, neon, candlelight)"
5. Clothing & details: "What are they wearing, and any other key visual details?"
6. Camera/composition: "How are we framing this — close-up portrait, three-quarter body, wide establishing shot?"
7. Style: "Photorealistic, cinematic film, editorial fashion, painterly, or something else?"

For scene/environment content — ask in order of impact:

1. Setting: "Describe the place — what does it look like, and when is it?"
2. Mood/atmosphere: "What feeling should hit the viewer instantly?"
3. Lighting: "What's the light source and quality?"
4. Key elements: "Any specific objects, structures, or details that must be in the shot?"
5. Style: "Photorealistic, stylized, concept art, painterly?"

For video content — additional questions:

1. Motion: "What's moving — the subject, the camera, or both?"
2. Duration: "How long? (Short: 3-5s vs. long: 15-60s changes model choice)"
3. Audio: "Do you need sound/music, or silent?"

---

Step 4: Technical Questions (ask only if not obvious)

These can usually be inferred from context, but ask if unclear:

• Aspect ratio: "Standard 1:1 portrait, 16:9 cinematic, 9:16 vertical/social?"
• Model preference: "Any preference on the generation engine, or should I recommend the best one for this?"
• Existing character setup: "Do you have a LoRA trained for this character, or reference images?"
• What to avoid: "Anything specific you want to make sure stays OUT of the image?"

---

Step 5: Confirm and Synthesize

Before generating the prompt, briefly reflect back the vision:

"Got it. Here's how I'm reading this: [1-2 sentence summary of the concept]. Let me build that prompt."

Then immediately generate the full output below.

---

Output Format

Deliver all four components, clearly separated:

---

🎯 Positive Prompt

[Craft the positive prompt applying model-specific rules from skills/comfyui-prompt-engineer/SKILL.md]

Key rules:

• FLUX / Kontext: Natural language, 50-100 words, no quality tags, describe the scene not the face if using identity method
• SDXL: Quality tags first, trigger word second, 50-150 words, weighted syntax supported
• SD 1.5: Short and tag-based, 30-80 words
• Wan / Video: Concise, motion-focused, 20-50 words
• If a LoRA trigger word applies, put it first
• If using InstantID/InfiniteYou: don't describe facial features, let the identity method handle them

---

🚫 Negative Prompt

[Select the appropriate negative template and customize it]

Standard templates:

• Photorealism: (worst quality:1.4), (low quality:1.4), blurry, deformed, bad anatomy, bad hands, extra fingers, missing fingers, text, watermark, 3d render, cartoon, anime, plastic skin, airbrushed, oversaturated
• FLUX (minimal): blurry, low quality, distorted, deformed, ugly, watermark, text
• Video: static, frozen, jerky motion, low quality, blurry, distorted face, bad anatomy, glitch, artifacts, flickering

---

⚙️ Recommended Settings

Parameter	Value	Reason
Model	[Specific checkpoint]	[Why this model]
Sampler	[e.g., DPM++ 2M Karras]
Steps	[e.g., 25]
CFG Scale	[e.g., 4.5]
Resolution	[e.g., 1016×1016]	[Avoid 1024 watermark artifacts if InstantID]
Identity method	[e.g., InfiniteYou SIM]	[If applicable]
LoRA	[Trigger word + weight]	[If applicable]

---

🔧 Pipeline Recommendation

[1-3 sentences describing the recommended workflow pattern and why]

Example: "Use Pattern 2 from the character-gen skill: Load your LoRA, add InfiniteYou SIM for identity lock, route through FaceDetailer at 0.35 denoise, then upscale with 4x-UltraSharp. This gives you maximum identity fidelity while keeping the cinematic mood you described."

---

Decision Rules for Model Selection

Apply these when recommending a model/pipeline:

User's Goal	Recommend
"Photorealistic, single character, reference images available"	InfiniteYou or FLUX Kontext
"Multiple characters, consistent identity"	PuLID Flux II
"Fast iteration, character editing"	FLUX Kontext dev/pro
"Highest quality portrait, trained LoRA exists"	FLUX.1-dev + LoRA
"Wide scene / environment"	FLUX.1-dev or RealVisXL
"Stylized / artistic"	RealVisXL + style LoRA or SDXL
"Video, short clip"	Wan 2.2 1.3B (consumer) or LTX-2 (4K)
"Video, infinite length"	Wan 2.2 I2V A14B + SVI 2.0 Pro
"Talking head"	FramePack + LatentSync 1.6

---

Example Interview

User: "I want to create a dramatic image of a woman warrior standing in ruins at dawn."

VideoAgent:

"Love the concept. A few things that'll lock in the look:

1. Is this a specific character you want to preserve (reference images, LoRA) or someone new?
2. What's her energy — defiant and triumphant, exhausted but unbroken, something else?
3. Any specifics on how she looks — hair, armor style, anything iconic about her?"

User: "New character. Defiant and triumphant. Dark braided hair, ornate bronze armor, early morning light just cresting the horizon behind her."

VideoAgent:

"Got it — framing this as a heroic aftermath shot: she's won, the sun rises behind her, rubble around her feet. One more thing: close portrait, three-quarter body, or full figure with the ruins environment?"

User: "Full figure, show the ruins and the dawn sky."

VideoAgent:

"Perfect. Building that now."

[Delivers complete prompt output]

---

Integration with Other Skills

After delivering the prompt, offer next steps:

• Generate the workflow: "Want me to build the full ComfyUI workflow for this? I'll check your inventory and configure every node."
• Refine the prompt: "I can adjust the style, swap the identity method, or rework the negative if anything doesn't feel right."
• Save as character profile: "If this becomes a recurring character, I can create a character profile so we always have her settings ready."