Sign In

Generate Detailed Images from Short Prompts with ERNIE Image in ComfyUI

0

Generate Detailed Images from Short Prompts with ERNIE Image in ComfyUI

You have an idea. A few words. You don't want to write a 200-word prompt to get a good image out of it.

Type a one-liner. Get a fully detailed image back.

Run it now on Floyo!

Why ERNIE Image

Most models need detailed prompts to produce detailed images. ERNIE Image has a built-in prompt enhancer a s,econd model that rewrites your short description into a richer visual prompt before generation runs.

Write one line. The enhancer fills in lighting, mood, composition, and detail. The image model gets a fully described scene. You get a detailed result without the prompt engineering.

Turn the enhancer off when you want your exact wording to drive the output instead.

  • built-in AI prompt enhancer, short prompts produce detailed results

  • enhancer reads your aspect ratio and shapes its description to match

  • toggle the enhancer off for precise prompt control

  • fast and forgiving, built for brainstorming and early concept work

  • 1024x1024 default, portrait and landscape aspect ratios supported

Key Inputs

Prompt

Your image description. With the enhancer on, a short idea is enough.

Short prompt examples that work well with the enhancer on:

  • "an abandoned Victorian mansion overtaken by vines, oil painting style"

  • "a lone astronaut on a red desert planet at dusk"

  • "cozy Japanese ramen shop at night, rain on the window, warm interior lighting"

  • "minimalist product shot of a glass perfume bottle, white background"

With the enhancer off, write a full descriptive prompt the way you would for any other model. Use it when you want precise control over wording, style, and composition without the enhancer rewriting your intent.

Enable Prompt Enhancement

Default: on. Keep it on for fast, low-effort generation. Turn it off when you need exact control over the output.

The tradeoff: enhancer on gives you more detail in exchange for some control over phrasing. Enhancer off gives you full control but requires a more complete prompt.

Negative Prompt

Leave empty for most prompts. Add specific exclusions if you're seeing recurring issues.

Common additions: "blurry, low quality, washed out colors, extra fingers, distorted"

Resolution

  • 1024x1024: default, works for most subjects

  • 832x1216: portrait orientation

  • 1216x832: landscape orientation

The enhancer reads your resolution and adapts its description to the aspect ratio automatically.

CFG: Default 4.

  • 2–3: looser, more creative interpretation

  • 5–6: tighter prompt adherence

Stay below 7. High CFG introduces burn and color artifacts on this model.

Steps: Default 20.

  • 12–16: faster previews

  • 25–30: more refinement on complex scenes

20 is the right balance for most use cases.

Seed: Randomize for variety. Fix a number to reproduce a result or compare prompt variations on the same composition.

What This Is Great For

Mood boards and early concepting: Type a rough idea, get a fully rendered visual direction back. Fast enough to generate a dozen variations in one session.

Illustration and concept art: Scene generation, fantasy environments, architectural concepts, and stylized illustration all work well with the enhancer active.

Non-technical users: The prompt enhancer removes the need to know prompt engineering. Describe what you want in plain language and the model handles the rest.

Rapid iteration: Fix a seed and change one word in your prompt to compare variations on the same composition. The enhancer produces consistent expansion logic across similar inputs.

What to Watch Out For

The enhancer rewrites your prompt. If precise wording matters, specific color values, exact composition instructions, a particular style you've carefully described, turn the enhancer off. Otherwise it will expand your input in its own direction.

This model handles scene generation and illustration well. For image editing and inpainting use a Klein 9B or Qwen Image Edit workflow. For strict character consistency across multiple shots, pair a Flux 2 Klein workflow with a Consistency LoRA instead.

CFG above 6 consistently introduces artifacts. Stay between 2 and 6 for clean output.

High step counts (30+) give diminishing returns on this model. 20 to 25 is the practical ceiling for most prompts.

0