Generation Flags

These options apply to generation. Every value defaults to unset: Bildomat sends no value unless you supply one or an adjustment derives one. --include-thoughts is off unless enabled. An omitted Boolean does not explicitly disable a provider default.

Supported flags and values depend on the model. Unsupported supplied options are omitted with a notice; --include-thoughts=false does not trigger an unsupported-option notice. Read parameter adjustment for replacement and omission rules, and commands for parsing and run-level flags.

Model Parameters

Flag and shorthand Type Meaning
--size, -s string Positive width and height, WxH. A valid supported size supersedes ratio and resolution.
--aspect-ratio, -a string Width-to-height ratio such as 16:9, or an allowed named value.
--resolution, -r string Resolution class, pixel height, or dimensions where supported. Derivation rules depend on the model.
--duration, -d integer Video length in seconds, from the values or range the model declares.
--quality, -q string Image quality level, from the values the model declares. A model declaring none passes the level to the provider as given.
--output-format, -f string Requested image format from the model's allowed values. A model declaring none passes the format to the provider as given. A supported image extension in output-path takes precedence.
--num-images, -N integer Number of images to generate, within the range the model declares.
--thinking-level, -l string Thinking level to apply to the model's generation effort, from the levels the model declares.
--include-thoughts, -n boolean Write model thought traces to a markdown file beside the artifact.
--input-media, -i string; repeatable File path or HTTP(S) URL of an image or video to edit or use as a reference.
--strength number How far the result may depart from the input image, from 0 (nearly identical) to 1 (minimal similarity).
--seed, -e integer Sampling seed, an integer, used by supporting providers.
--background string Background treatment of the generated image, from the treatments the model declares.
--output-compression integer Compression setting for JPEG or WebP. Bildomat sends the value for any output format, including PNG; the provider decides how to apply it.
--safety-tolerance integer Moderation strictness within the range the model declares, where 0 is strictest and higher numbers are more permissive.
--generate-audio boolean Generate a soundtrack alongside the video.
--prompt-upsampling boolean Let the provider expand the prompt before generating.
--disable-prompt-upsampling boolean Turn off the provider's automatic prompt expansion.
--moderation-level string Content moderation strictness applied by the provider, from the levels the model declares.
--person-generation string Whether the model may generate people, and of which ages, as a policy word the model's page explains.
--negative-prompt string Text describing what to exclude from an image or video.
--guidance-scale number Strength of prompt guidance, within the range the model declares.
--steps integer Number of generation steps, within the range the model declares.
--expand-top integer Pixels to add above the input image.
--expand-bottom integer Pixels to add below the input image.
--expand-left integer Pixels to add to the left of the input image.
--expand-right integer Pixels to add to the right of the input image.
--finetune-id string ID of a model customized through provider training. It must already be owned by or shared with your account.
--finetune-strength number How strongly to apply the selected fine-tune.
--mask string Encoded mask data or a mask URL, as supported by the model.
--mask-dilation integer Pixels by which to expand the masked area.
--garment-url string Public HTTP(S) URL of the garment image for virtual try-on.
--style-id string Identifier of an existing provider style compatible with the selected model.
--style-match string How closely the generated image follows the selected style, for example precise or flexible.
--watermark boolean Include the provider watermark in the generated video.
--upscale-factor number Multiplier applied to the input video dimensions.
--creativity integer Amount of detail the provider may invent while upscaling.
--voice-id string Provider voice identifier used to speak the prompt, for example 16a09e4706f74997ba4ed05ea11470f6.
--motion-prompt string Text describing the avatar's gestures and movement.
--expressiveness string Avatar expression and movement intensity, for example low, medium, or high.
--image-fit string How the avatar image fits the output canvas, for example contain or cover.
--remove-background boolean Remove the original background from the avatar image.
--voice-speed number Playback speed multiplier for generated speech.
--voice-pitch number Pitch adjustment in semitones.
--voice-volume number Volume of generated speech, from silent to full volume.
--voice-locale string Locale or accent hint for a multilingual voice, for example en-US or en-GB.

String flags take text; quote values containing spaces. Integer flags take whole numbers, number flags take numerical values, and Boolean flags use --flag or --flag=false. Invalid numerical syntax exits 2; nonfinite numbers that parse are omitted with a notice.

--input-media is the only repeatable model option. Repeat it for URLs; comma-separated local paths also work. See input media for formats, timing prefixes, validation, and count limits.

--strength ranges from 0 (little change) to 1 (minimal similarity) on direct Recraft editing models. Supply it when editing; it is optional for generation without an image. On bfl/flux-pro-1.1-ultra-finetuned, --strength accepts 0 to 1 and applies to the optional reference image. --seed does not guarantee an identical result on a later request.

--mask and --garment-url send their text to the provider unchanged. Bildomat does not read a local file named there, so supply a URL, or encoded data where the model accepts it. Only --input-media reads local files.

Transparent raster output requires a format with an alpha channel, such as PNG or WebP. Background values and compression combinations remain subject to the provider's checks. --prompt-upsampling and --disable-prompt-upsampling are separate options declared by different models; neither is a universal switch.

Models Accepting Each Flag

Each provider link lists the exact model options, allowed values, ranges, required inputs, and conditional rules. "All" means every model of that provider; counts describe the catalog covered by these pages.

Flag Provider coverage
--size OpenAI: all; OpenRouter: 1 of 90 models; Black Forest Labs: 10 of 25 models; Recraft: 11 of 21 models
--aspect-ratio OpenAI: all; xAI: all; OpenRouter: 85 of 90 models; Google: all; Black Forest Labs: 15 of 25 models; Sourceful: all; Recraft: all; Kling: all
--resolution OpenAI: all; xAI: all; OpenRouter: 75 of 90 models; Google: 8 of 10 models; Black Forest Labs: 11 of 25 models; Sourceful: all; Recraft: 11 of 21 models; Kling: all
--duration OpenAI: 2 of 9 models; xAI: 3 of 7 models; OpenRouter: 27 of 90 models; Google: 3 of 10 models; Black Forest Labs: 1 of 25 models; Kling: 6 of 10 models
--quality OpenAI: 7 of 9 models; xAI: 1 of 7 models; OpenRouter: 44 of 90 models
--output-format OpenAI: 7 of 9 models; OpenRouter: 46 of 90 models; Black Forest Labs: 21 of 25 models; Sourceful: 2 of 4 models; Recraft: 1 of 21 models
--num-images OpenAI: 7 of 9 models; xAI: 4 of 7 models; OpenRouter: 55 of 90 models; Recraft: all; Kling: 4 of 10 models
--thinking-level Google: 4 of 10 models; Sourceful: 2 of 4 models
--include-thoughts Google: 4 of 10 models
--input-media OpenAI: all; xAI: all; OpenRouter: 87 of 90 models; Google: all; Black Forest Labs: 22 of 25 models; Sourceful: all; Recraft: 14 of 21 models; Kling: all
--strength Black Forest Labs: 1 of 25 models; Recraft: 14 of 21 models
--seed OpenRouter: 31 of 90 models; Google: 3 of 10 models; Black Forest Labs: 20 of 25 models; Recraft: all
--background OpenAI: 7 of 9 models; OpenRouter: 44 of 90 models; Sourceful: 2 of 4 models
--output-compression OpenAI: 7 of 9 models; OpenRouter: 44 of 90 models
--safety-tolerance OpenRouter: 3 of 90 models; Black Forest Labs: all
--generate-audio OpenRouter: 18 of 90 models; Black Forest Labs: 1 of 25 models; Kling: 3 of 10 models
--prompt-upsampling Black Forest Labs: 10 of 25 models; Sourceful: all
--disable-prompt-upsampling Black Forest Labs: 4 of 25 models
--moderation-level OpenAI: 7 of 9 models
--person-generation Google: 3 of 10 models
--negative-prompt Google: 2 of 10 models; Recraft: 4 of 21 models; Kling: 2 of 10 models
--guidance-scale Black Forest Labs: 5 of 25 models
--steps Black Forest Labs: 5 of 25 models
--expand-top Black Forest Labs: 1 of 25 models
--expand-bottom Black Forest Labs: 1 of 25 models
--expand-left Black Forest Labs: 1 of 25 models
--expand-right Black Forest Labs: 1 of 25 models
--finetune-id Black Forest Labs: 2 of 25 models
--finetune-strength Black Forest Labs: 2 of 25 models
--mask Black Forest Labs: 2 of 25 models
--mask-dilation Black Forest Labs: 1 of 25 models
--garment-url Black Forest Labs: 2 of 25 models
--style-id Recraft: 4 of 21 models
--style-match Recraft: 4 of 21 models
--watermark OpenRouter: 1 of 90 models
--upscale-factor OpenRouter: 1 of 90 models; Black Forest Labs: 1 of 25 models
--creativity OpenRouter: 1 of 90 models; Black Forest Labs: 1 of 25 models
--voice-id OpenRouter: 1 of 90 models
--motion-prompt OpenRouter: 1 of 90 models
--expressiveness OpenRouter: 1 of 90 models
--image-fit OpenRouter: 1 of 90 models
--remove-background OpenRouter: 1 of 90 models
--voice-speed OpenRouter: 1 of 90 models
--voice-pitch OpenRouter: 1 of 90 models
--voice-volume OpenRouter: 1 of 90 models
--voice-locale OpenRouter: 1 of 90 models

For a machine-readable answer for a selected model, use bild info PROVIDER/MODEL --json. Every accepted flag appears in that model's params; general definitions appear in flags.

Bildomat 0.0.7Revised 2026-10-13