🧬 AI Influencer Portrait Generator — Zero LoRA, Pure SDXL

Generate ultra-realistic, photographic AI influencer portraits in seconds. No LoRA, no custom nodes — just drop the JSON into ComfyUI and hit Queue.

"I built this to generate consistent, publication-quality influencer portraits for social media campaigns. After dozens of iterations, this is the cleanest SDXL setup I've found for photorealistic faces."

⚙️ Workflow Specs

SettingValueBase ModelSDXL Base 1.0Resolution768 × 1024 (portrait)SamplerDPM++ 2M · KarrasSteps / CFG28 steps · CFG 7Upscaler4x-UltraSharpBatch Size6 images per runSeedFixed 123456 (change to Random for variation)Custom Nodes❌ None required

🔗 Node Pipeline

01 — CheckpointLoaderSimple Loads sd_xl_base_1.0.safetensors. No refiner needed.

02 — CLIPTextEncode (Positive) Crafted influencer prompt: Nordic features, studio fashion photography, 85mm shallow DOF, authentic skin texture.

03 — CLIPTextEncode (Negative) Targets blur, anatomy errors, plastic skin, anime artifacts, and over-sharpening.

04 — EmptyLatentImage 768×1024 canvas, batch 6 — optimized for Instagram & TikTok formats.

05 — KSampler DPM++ 2M Karras, 28 steps, CFG 7. Fixed seed for reproducibility — switch to Random for unique faces.

06 — VAEDecode Decodes latent to pixel using the bundled SDXL VAE.

07 — UpscaleModelLoader + ImageUpscaleWithModel 4x-UltraSharp upscaler — gold standard for photorealistic faces. Final output ≈ 3072×4096px.

08 — SaveImage Saves all 6 upscaled images as PNG to your ComfyUI output folder.

📝 Included Prompts

Positive:

portrait photo of a Scandinavian woman, authentic Nordic facial features,
blonde hair, clear blue eyes, fair skin, elegant symmetrical face,
sharp intelligent gaze, defined cheekbones, natural lips,

ultra realistic AI influencer, instagram model,
fashion photography, minimalist modern outfit,

soft nordic lighting, cool color tones,
studio portrait, professional photography,

extremely detailed skin texture, natural pores, photorealistic,
85mm lens, shallow depth of field, high dynamic range,

solo, centered composition, looking at camera,
clean blurred background, instagram lifestyle aesthetic

Negative:

low quality, blurry, bad anatomy, deformed face,
extra fingers, extra limbs, distorted eyes,
cartoon, anime, painting, illustration,
plastic skin, oversharpen, overexposed

💡 Pro Tips

🎲 Change KSampler seed from Fixed → Random for unique face variations each run.

🌍 Swap "Scandinavian woman" for any ethnicity — Japanese, Brazilian, Middle Eastern — and adjust hair/eye descriptors.

📷 Add LEICA M11, Fujifilm XT5, Sony A7RV to the positive prompt for a more authentic camera-captured feel.

💡 Swap "soft nordic lighting" → "golden hour sunlight" or "neon city lights" for different moods.

⚡ Pair with IP-Adapter Face ID nodes to lock a face and generate a consistent influencer series.

🔁 Feed outputs into img2img at denoise 0.35–0.45 to refine details while keeping the composition.

📋 Requirements

→ ComfyUI v0.12.3 or newer → sd_xl_base_1.0.safetensors in models/checkpoints/ → 4x-UltraSharp.pth in models/upscale_models/ → 8 GB VRAM minimum (12 GB+ recommended for batch of 6) → No custom nodes required ✅

📝 Changelog

V01 — Initial release. SDXL pipeline with 4x-UltraSharp upscaling, batch 6, curated influencer prompts. Zero custom node dependencies.

V02 (coming soon) — IP-Adapter Face ID integration for consistent character generation across sessions.