Short answer
Choose by output type first, then by resolution, reference-image needs, speed, and how the selected settings affect credits. The newest choices are Nano Banana Lite, Seedream 5.0 Pro, Veo 3.1 Lite, and Seedance 2.0 Mini.
Image model overview
• Default: Grok Imagine Image Quality at 1K and 3:4.
• Grok Imagine models: straightforward image creation at 1K or 2K with up to three references.
• Nano Banana Lite: 1K only with up to 14 references.
• Nano Banana 2 and Pro: flexible Gemini image options with up to 14 references and higher output resolutions.
• GPT Image 2: text-to-image and image-to-image with an estimated charge based on medium quality.
• Seedream 4.5 and Seedream 5.0 Lite: higher-resolution choices. Seedream 5.0 Lite supports up to 4K and 14 references.
• Seedream 5.0 Pro: 1K or 2K, defaults to 2K, and accepts up to 10 references.
Video model overview
• Default: Veo 3.1 Fast at 720p, 9:16, and 8 seconds.
• Veo 3.1 and Veo 3.1 Fast: 720p, 1080p, or 4K. Veo 3.1 Lite offers 720p or 1080p.
• Grok Imagine Video: a one-image workflow with 4-, 8-, or 12-second options. Video 1.5 supports up to 1080p and requires an image.
• Seedance 2.0: up to 4K and as many as nine references.
• Seedance 2.0 Fast and Seedance 2.0 Mini: 480p or 720p for faster drafts and up to nine references.
• Seedance 1.5 Pro: up to 1080p, with reference limits that depend on the chosen resolution.
Settings shared by new videos
New video generations support only 9:16 or 16:9. Non-Veo models offer 4, 8, or 12 seconds. Veo offers 4 or 8 seconds at 720p and requires 8 seconds at higher resolutions. Adaptive ratio and automatic duration are unavailable for new generations.
Credits
Image charges apply per generated image. Video charges apply once per configured clip. Charges depend on the model, resolution, duration, and accepted reference count. A retry, edit, or append starts a new charged attempt. GPT Image 2 uses an estimate that is not adjusted after completion.
Troubleshooting
• If a setting disappears after changing models, choose one supported by the newly selected model.
• If you are comparing quality, keep the prompt and main settings consistent across tests.
• Use a Lite, Fast, or Mini model for lower-resolution drafts, then move to a higher-quality model for final output.