TokiToki Internal
Atlas + S3
All runs
CompletedComfyUI

e49943d8-d84c-4402-89a3-07455c36494a

Queued
Jul 22, 2026, 03:43 PM
Completed
Jul 22, 2026, 03:43 PM
Execution time
13.8s
Outputs
1 · 1.3 MB
Inputs
3
Cached nodes

Outputs

1

Click an image to view it full size.

character_ref_00363_.png

node 3 · 1.3 MB

Text prompts

3
Promptnode 1 · PrimitiveStringMultiline · valueCharacter Description (your input)

[REFS: attach @Elena and @Aidan face refs] Low angle shot from between and below the two people on the porch steps — camera is at step level looking UP at both of them. The woman is upper-left in frame looking down toward the man. The man is upper-right in frame looking up toward her. The grey overcast sky fills the space between and above their heads. The porch railing posts frame them on either side. The two mugs are visible in the extreme foreground, large and soft. Both faces are seen from below — jaw lines, under-chins, the vulnerability of faces seen from an angle people never present to each other. The sky between their heads is the gap — it will close. Shot on full-frame mirrorless, 28mm at f2.8 from low. Both faces readable but the wide angle from below creates slight distortion giving intimacy an edge of unease. Natural overcast light from the sky above filling their faces evenly from overhead. Hyperrealistic. 9:16 vertical.

Systemnode 2 · GeminiImageDirect · system_promptCharacter Image Generator (Nano Banana)

Generate a single, hyper-realistic character reference image (strictly no collages, text, watermarks, or borders). The image must feature a cinematic, three-quarter portrait of the described character, maintaining a sharp, tack-focused emphasis on the face with natural skin textures, pore details, and realistic micro-expressions. The lighting should be cinematic and directional (e.g., subtle split or Rembrandt lighting) to give depth, blending seamlessly with the environment specified in the character description. The character must have a natural, living presence, looking slightly off-camera or directly at the lens with a subtle, context-aware expression. Ensure the full head, neck, and upper shoulders are completely unobstructed. This image serves as a high-fidelity identity blueprint for temporal-consistent video generation.

1 other text field— sigmas, filenames, and other non-prompt strings the logger swept up

node 5 · LoadImage · image

WhatsApp Image 2026-07-21 at 11.47.35 PM.jpeg

Input media

3

Source images and video this run consumed.

character_ref_00360_.png

node 4 · 1.4 MB

WhatsApp Image 2026-07-21 at 11.47.35 PM.jpeg

node 5 · 99 KB

character_ref_00348_.png

node 6 · 1.4 MB