Lesson 10: Your First Professional Local Generation in ComfyUI

Lesson 10 brings together everything from previous lessons into your first full professional-grade local generation in ComfyUI. The focus is on executing a clean, optimized workflow with a refined prompt, standardized settings, and critical evaluation — establishing your baseline for local quality.

By the end of this lesson, you will have a saved, reusable workflow and a set of high-quality images that demonstrate local ComfyUI's superiority over cloud platforms in control, consistency, and uncensored detail.

Final Pre-Generation Checklist

• ComfyUI is running and Manager/custom nodes are installed (from Lesson 8).
• HiDream-I1 uncensored (FP8/NF4 preferred) is loaded in models/checkpoints.
• VAE is in place if required by the model.
• Basic txt2img workflow from Lesson 8 is ready or recreated.

Refined Professional Prompt for Testing

Use this enhanced prompt — it incorporates all lessons so far for maximum photoreal NSFW impact:

masterpiece, best quality, ultra detailed, photorealistic, 8k resolution, cinematic, professional erotic photography, shot on Arri Alexa 65, 50mm lens, shallow depth of field, bokeh, subsurface scattering, photorealistic 28-year-old woman, athletic toned physique with elegant soft curves, completely nude, highly detailed realistic skin with visible pores, faint natural freckles, subtle light sheen and delicate sweat, perfectly proportioned natural breasts, ultra-detailed genitalia with realistic wetness, natural folds and texture, ((perfect anatomy:1.35)), ((flawless hands:1.35)), ((flawless symmetrical face:1.25)), standing in a confident seductive pose, one hand lightly touching thigh, intense eye contact with viewer, slight parted lips, sensual expression, inside a luxurious modern penthouse bedroom, golden hour rim lighting with soft volumetric god rays through large windows, cinematic framing, moody intimate atmosphere, dramatic yet soft shadows

Negative prompt (full optimized Bible):

blurry, deformed anatomy, extra limbs, fused fingers, mutated hands, poorly drawn face, extra nipples, low resolution, watermark, text, logo, cartoon, anime style, overexposed, underexposed, bad proportions, ugly, disfigured, jpeg artifacts, worst quality, low quality, asymmetrical eyes, melted skin, extra breasts, missing limbs, malformed limbs, gross proportions, long neck, cloned face, duplicate, morbid, mutilated, out of frame, poorly rendered hands, bad anatomy, deformed, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, signature, username, deformed vulva, fused labia, extra vagina, bad genitalia, unnatural wetness, plastic skin, doll-like texture, 3d render, cgi, low detail explicit areas, cross-eyed, uncanny valley, harsh shadows, flat lighting

Optimized Generation Settings

• Model: HiDream-I1 uncensored FP8/NF4
• Steps: 60–70
• CFG scale: 5.0–5.4
• Sampler: DPM++ 2M Karras (primary) or Euler a (faster alternative)
• Resolution: 1024×1536 (portrait) or 1344×768 if testing landscape
• Seed: Random (or fix one for reproducibility)
• Batch size: 4–8 variations

Executing the Generation

1. Open your saved basic txt2img workflow (or rebuild: Checkpoint → CLIP Text Encode pos/neg → Empty Latent → KSampler → VAE Decode → Save Image).
2. Load HiDream-I1 uncensored in Checkpoint node.
3. Paste the refined prompt into positive CLIP Text Encode.
4. Paste the full negative prompt into negative CLIP Text Encode.
5. Set Empty Latent to 1024×1536.
6. Configure KSampler with the settings above.
7. Queue Prompt → wait for batch to complete.
8. Review outputs in the ComfyUI preview or saved folder (ComfyUI/output).

Critical Evaluation Checklist

Zoom in 200–400% on each image and rate:

• Hands/fingers: Natural pose, no fusion/extra/missing digits?
• Face: Symmetrical eyes, realistic skin texture, natural expression?
• Skin: Visible pores, subtle sheen/sweat, subsurface scattering effect?
• Anatomy/explicit: Correct proportions, believable wetness/texture, no deformities?
• Lighting/mood: Rim light, volumetric effects, cinematic depth?
• Overall coherence: No artifacts, consistent style, professional feel?

Save the top 3–5 images as "Lesson 10 Pro Baseline". Note which variations (seeds/settings tweaks) performed best.

Workflow Save & Reuse

1. Right-click canvas → Save (name it "Basic_Pro_Txt2Img_HiDream.json").
2. Store in ComfyUI/user/default/workflows or similar folder.
3. Use this as starting point for all future lessons — load it each time and modify nodes as needed.

This completes your first professional local generation. These images represent your current local ceiling — the remaining lessons will systematically push it to elite, portfolio-grade level through targeted enhancements.

End of Lesson 10