What this flow does
This is a do-it-all finisher for reference-driven shoots. For each uploaded reference image, we:
Analyze the photo (pose, framing, lighting, styling) and generate similar—but not identical—shots.
Upscale the results for detail and clarity (you choose Basic for speed or Premium for higher realism).
Face Swap every image to the same source face so the identity is consistent across the entire batch.
Use it when you know the look (reference) and the face (Persona) you want—and you need finished images fast.
When to use this flow
You found a great reference (Pinterest/IG/your own photo) and want a near-match pose/location/mood with variations.
You’re producing a campaign batch and want identity locked across 10–1000 images.
You prefer one click instead of hopping between tools (Gen by Ref → Upscale → Face Swap).
If you still plan heavy edits (dress-up, background replace), do those first and run this flow last.
Inputs (what each control means)
Upload Reference Images
Add up to 1000 references. Each reference yields four variations. Use sharp, well-lit images that clearly show pose and framing.Positive Prompt (optional)
Add what you want to carry over: “soft window light, shallow depth of field, neutral grade, tidy background”.Negative Prompt (optional)
Ban artefacts or scene elements: “flicker, plastic skin, heavy sharpening, messy background, watermark, text”.Strength (slider)
How closely we follow the reference. Higher = closer to original pose/composition; lower = freer interpretation.LoRAs (optional)
Apply curated SDXL LoRAs (e.g., physique/style packs). Combine sparingly; higher strengths can introduce artefacts.Upload a Face Image
The identity we’ll put on every generated image. Use a sharp, front or slight 3/4 portrait, no sunglasses, minimal occlusion.Upscaler Version
Basic = faster/cheaper; Premium = more realism/micro-detail (slower/costlier). Face Swap quality is identical in both; this choice only affects the upscale.Comment
Name your job for easier tracking (“ref-beach-campaign-aug” etc.).
Quick start (3 minutes)
Upload references (1–1000).
Set Strength around the middle; add a (optional) concise positive and a tight negative.
(Optional) Pick LoRAs for physique/styling.
Upload your Face Image (Persona).
Choose Upscaler: Basic (speed) or Premium (finals).
Start. You’ll get 4 variations per reference, already upscaled and face-swapped.
Pro tips for high-quality, consistent results
Curate references: clean edges, clear pose, simple background. One reference per pose is better than a collage.
Keep motion simple: dramatic hair across the face or heavy occlusions can hurt swap quality.
Batch smart: group references by scene (beach set, office set). Consistent lighting = easier identity match.
LoRAs: mix at most 1–2; start with moderate strength and only increase if you truly need more influence.
Premium for finals: use Basic to explore, Premium to finish your keepers.
Naming: add persona/scene/ratio to job names for downstream organization.
Common issues & fixes
Identity looks off/uncanny → Use a sharper, front-lit face source; avoid sunglasses; raise reference Strength a bit so pose/lighting stabilize.
Skin tone mismatch → Prefer Premium upscale (better tone harmonizing) and choose references with similar lighting to your face source.
Artefacts with multiple LoRAs → Reduce LoRA strengths or use only one; higher strengths increase artefact risk.
Too close to the reference (not enough variation) → Lower Strength slightly and add a few prompt hints (e.g., “wider lens feel”, “different hand placement”).
FAQ
How many outputs do I get?
Four per reference (fixed for now). Uploading 25 references yields 100 finished images.
Can I mix different faces in one run?
No—one source face per job for maximum consistency. Run separate jobs for different Personas.
Do LoRAs work here?
Yes (SDXL). Use them to guide physique or style. We auto-inject tokens; keep prompts human-readable.
What does the Upscaler choice change?
Only the enhancement step. Basic is fast/economical; Premium adds realism/micro-detail. Face Swap quality is the same.
Is this safe for products/e-com?
If you need zero change beyond resolution, use Sharp Keep outside this flow. This flow prioritizes identity locking + photographic realism.
Why creators love this flow
It replaces a 3-step pipeline with one job, producing cohesive, high-detail images where the same face appears across every shot. Perfect for campaigns, lookbooks, and weekly content packs.