Loop ForgeOne reference, six scenes
From Qwen-Image 2.1 Turbo, run locally
- Output
- 832×1216
- Turbo, 8 steps
- 23.3 to 24.1 s per image
- Base 2.1, 40 steps
- 62.9 to 63.5 s per image
- Turbo sampling
- 8, euler, trained sigmas (ManualSigmas), cfg 1.0
- Base sampling
- 40, euler / simple, cfg 1.0
- Seed
- 21
- Ran on
- RTX 3060, 12GB (local)
What was wired in
Place the person from the image in a busy city street at night, lit by shop signs, standing on the pavement and looking toward the camera, framed from the waist up.
Place the person from the image on a windswept coastal cliff path in daylight, framed from the waist up, with the sea far behind and below.
Place the person from the image at a workbench in a bright workshop, wearing the same clothing, holding a small hand tool and looking up toward the camera.
Place the person from the image in a warmly lit library aisle between tall shelves, framed from the waist up, one hand resting on a shelf.
Place the person from the image in a commercial kitchen in daylight, framed from the waist up, standing beside a stainless steel counter.
Place the person from the image on a railway platform in cold morning light, framed from the waist up, with the platform receding behind.
Notes
One reference photo and a short prompt per scene. Identity held in all six scenes on both models.
Workflow: Turbo, 8 steps, cliff path
Workflow: Turbo, 8 steps, workshop
Workflow: Turbo, 8 steps, library
Workflow: Turbo, 8 steps, kitchen
Workflow: Turbo, 8 steps, platform
Workflow: Base 2.1, 40 steps, city street
Workflow: Base 2.1, 40 steps, cliff path
Workflow: Base 2.1, 40 steps, workshop
Workflow: Base 2.1, 40 steps, library