ERNIE-Image Trainer
ERNIE-Image Trainer generates vivid, style-consistent visuals from text prompts.
A tradução deste card ainda não está disponível; parte do conteúdo aparece em inglês.
ERNIE-Image Trainer is a text-to-image model fine-tuned to produce sharp, artistic, and photorealistic results across many styles. It excels when you need consistent characters, coherent lighting, or specific artistic movements in seconds. The model is distinctive for its balanced detail and speed, making it ideal for iterative design work, concept exploration, and rapid prototyping without leaving your creative workspace.
- character concept art
- social media banners
- product mockups
- illustration style transfer
- storyboard scene generation
Maintains requested artistic styles and color palettes with high accuracy.
Delivers usable images in seconds, supporting quick creative experiments.
Produces intricate textures and lighting details even in complex scenes.
- Start with a clear subject and style keyword in your prompt
- Use negative prompts to remove unwanted elements
- Adjust aspect ratio early to avoid cropping surprises
- Generate multiple seeds for variety when exploring ideas
- May struggle with highly abstract or surreal concepts not seen in training data
- Occasionally introduces minor artifacts in fine textures or edges
- Limited to generating single images; no native video or 3D output
Majestic floating castle above clouds at sunset, epic fantasy art, glowing windows, dramatic sky, highly detailed, wide cinematic aspect ratio
Portrait of an elderly scholar in a dimly lit library, oil painting style, warm lamplight, leather-bound books, intricate robe textures, Rembrandt lighting
Vibrant cosmic nebula in deep space, abstract digital art, swirling purples and teals, starfield backdrop, glowing gas clouds, ultra high resolution
Random starting point for generation; same seed with same prompt produces identical results.
Number of refinement iterations; more steps usually mean finer detail but take longer.
Width-to-height ratio of the output image, e.g., 16:9 for widescreen or 1:1 for square.
How strictly the image follows your prompt; higher values increase prompt adherence but may reduce creativity.
Os preços exibidos são o que você paga na Muitto. Sem surpresas.
What file formats are supported for output?
Outputs are delivered as standard PNG or JPEG files ready for immediate use.
Can I train on my own custom dataset?
The model is pre-trained; fine-tuning on custom data requires separate tooling.
Is there a limit on prompt length?
Prompts up to 75 tokens work best; longer prompts may be truncated.