A SMALL TEXT-TO-IMAGE RECIPE

Checkpoint: Lykon / DreamShaper_8_pruned.safetensors
Base family: Stable Diffusion 1.5
App: Stability Matrix -> ComfyUI
Size: 768 x 512
Seed: 101509
Steps: 28
CFG: 6.5
Sampler: dpmpp_2m
Scheduler: karras
Batch size: 1

PROMPT
Wide cinematic illustration of a tiny coastal village built around a sheltered harbour, warm amber lanterns along a stone quay, small wooden boats floating on calm blue water, indigo dusk sky, cozy glowing windows, painterly detailed environment concept art, clear architecture, no people, no writing

NEGATIVE PROMPT
text, watermark, logo, signature, blurry, low quality, distorted, duplicate, oversaturated, nude, nudity, explicit

NODE PATH
CheckpointLoaderSimple -> CLIPTextEncode (positive and negative) -> EmptyLatentImage -> KSampler -> VAEDecode -> SaveImage

WHAT TO COMPARE
Keep the seed and settings fixed, then change only “indigo dusk sky” to “soft sunrise sky.” Compare atmosphere, architecture and boats. Fixed settings improve the comparison, although results can still differ across software versions and hardware.

This recipe was actually run locally for the harbour image used in our short-video example. The video adds an edited camera move to a single generated still; it is not a video diffusion result.