7466119867
* feat(image): 图片参数 - 统一插件声明配置与生成合同 * docs(image): 图片参数 - 说明声明合同与验证边界 * fix(plugins): 图片声明 - 保留内联素材支持并同步制品文档 * fix(image): 参数合同 - 对齐审批意图和尺寸预检调用方 * fix(image): 调用链 - 保留公开意图并接通独立分辨率 * fix(image): 图片参数 - 补齐分辨率接线与目录价格投影
221 lines
12 KiB
Go
221 lines
12 KiB
Go
// 各类模型的默认能力配置与按模型的特例修正（例如固定分辨率、特定时长档位）。
|
||
|
||
package app
|
||
|
||
import (
|
||
"strconv"
|
||
"strings"
|
||
|
||
"yingce/backend/internal/model"
|
||
)
|
||
|
||
func DefaultModelCapabilityConfig(protocol string) *ModelCapabilityConfig {
|
||
return DefaultModelCapabilityConfigForModel(protocol, "")
|
||
}
|
||
|
||
func videoDurationSupported(value *VideoCapabilityConfig) bool {
|
||
return value == nil || value.DurationSupported == nil || *value.DurationSupported
|
||
}
|
||
|
||
func DefaultImageCapabilityConfig(protocol string, modelName string) *ImageCapabilityConfig {
|
||
image := &ImageCapabilityConfig{
|
||
References: ImageReferenceConfig{PromptMaxChars: 32000, MaxImages: 16, MaxImageBytes: 30 * 1024 * 1024, MaskSupported: true},
|
||
Size: ImageSizeConfig{Parameter: "size", Values: defaultImageSizeValues(), Default: "1:1", AllowCustom: true},
|
||
Quality: ImageQualityConfig{Supported: true, Values: []string{"auto", "low", "medium", "high"}, Default: "auto"},
|
||
TransparentBackground: VideoBooleanConfig{Supported: true, Default: false},
|
||
ResponseFormat: ParameterSupport{Supported: true},
|
||
OutputFormat: ParameterSupport{Supported: true},
|
||
MaxOutputs: 15,
|
||
}
|
||
switch model.ChannelInterfaceType(protocol) {
|
||
case model.ChannelInterfaceOpenAIImage:
|
||
// OpenAI Images 只收 WIDTHxHEIGHT：默认配置声明了 1K/2K/4K，就必须自带档位换算表，
|
||
// 否则运行时换算会被静默跳过（见 imageSizePresetMissing），实际出图档位与所选档位不一致。
|
||
image.Size.Presets = openAIImageDefaultSizePresets()
|
||
case "km-kemei-seedream":
|
||
image.References.MaxImages = 10
|
||
image.References.MaskSupported = false
|
||
presets := kemeiSeedreamImagePresets()
|
||
values := make([]string, 0, len(presets))
|
||
for _, preset := range presets {
|
||
values = append(values, preset.Size)
|
||
}
|
||
image.Size = ImageSizeConfig{Parameter: "size", Values: values, Default: "2048x2048", AllowCustom: true, Presets: presets}
|
||
image.Quality = ImageQualityConfig{Supported: false, Values: []string{}, Default: "auto"}
|
||
// 透明背景需要带 alpha 的单张参考图，通过插件命名空间显式配置。
|
||
image.TransparentBackground = VideoBooleanConfig{Supported: false, Default: false}
|
||
image.MaxOutputs = 1
|
||
case model.ChannelInterfaceGrokImage:
|
||
image.References.MaxImages = 1
|
||
image.References.MaskSupported = false
|
||
// grok2api / xAI Imagine：size→aspect_ratio，quality→resolution(1k/2k)。
|
||
image.Size = ImageSizeConfig{Parameter: "aspect_ratio", Values: []string{"1:1", "3:4", "4:3", "9:16", "16:9", "2:3", "3:2"}, Default: "1:1", AllowCustom: false}
|
||
image.Resolution = ImageResolutionConfig{Supported: true, Values: []string{"1k", "2k"}, Default: "2k"}
|
||
image.Quality = ImageQualityConfig{Supported: false, Default: "auto"}
|
||
image.TransparentBackground = VideoBooleanConfig{Supported: false, Default: false}
|
||
image.ResponseFormat = ParameterSupport{Supported: true}
|
||
image.OutputFormat = ParameterSupport{Supported: false}
|
||
image.MaxOutputs = 1
|
||
case model.ChannelInterfaceVolcengineArkImage, model.ChannelInterfaceVolcengineArkAgentPlanImage:
|
||
image.References.MaskSupported = false
|
||
image.Quality.Supported = false
|
||
image.TransparentBackground.Supported = false
|
||
image.ResponseFormat.Supported = false
|
||
image.OutputFormat.Supported = false
|
||
case model.ChannelInterfaceVolcengineJiMengImage:
|
||
image.References.MaxImages = 14
|
||
image.References.MaskSupported = false
|
||
image.Quality.Supported = false
|
||
image.TransparentBackground.Supported = false
|
||
image.ResponseFormat.Supported = false
|
||
image.OutputFormat.Supported = false
|
||
case model.ChannelInterfaceGeminiImage:
|
||
image.References.MaskSupported = false
|
||
// Gemini Images uses imageConfig.aspectRatio, not the OpenAI-style pixel size field.
|
||
image.Size = ImageSizeConfig{Parameter: "aspect_ratio", Values: []string{"auto", "1:1", "2:3", "3:2", "3:4", "4:3", "9:16", "16:9", "21:9"}, Default: "1:1", AllowCustom: false}
|
||
image.TransparentBackground.Supported = false
|
||
image.ResponseFormat.Supported = false
|
||
image.OutputFormat.Supported = false
|
||
image.MaxOutputs = 4
|
||
image.ResolutionUsesQuality = true
|
||
}
|
||
if model.ChannelInterfaceType(protocol) != model.ChannelInterfaceGrokImage && strings.HasPrefix(strings.ToLower(strings.TrimSpace(modelName)), "grok-imagine-image") {
|
||
image.References.MaxImages = 0
|
||
image.References.MaskSupported = false
|
||
image.Size = ImageSizeConfig{Parameter: "aspect_ratio", Values: []string{"1:1", "3:4", "4:3", "9:16", "16:9", "2:3", "3:2"}, Default: "1:1", AllowCustom: false}
|
||
image.Resolution = ImageResolutionConfig{Supported: true, Values: []string{"1k", "2k"}, Default: "2k"}
|
||
image.Quality = ImageQualityConfig{Supported: false, Default: "auto"}
|
||
image.TransparentBackground = VideoBooleanConfig{Supported: false, Default: false}
|
||
image.ResponseFormat = ParameterSupport{Supported: true}
|
||
image.OutputFormat = ParameterSupport{Supported: false}
|
||
image.MaxOutputs = 1
|
||
}
|
||
return image
|
||
}
|
||
|
||
func defaultImageSizeValues() []string {
|
||
return []string{
|
||
"auto", "1:1", "3:2", "2:3", "4:3", "3:4", "16:9", "21:9", "9:16",
|
||
"1024x1024", "1360x1024", "1024x1360", "1536x1024", "1024x1536", "1024x1280", "1280x1024", "2048x878", "1824x1024", "1024x1824",
|
||
"2048x2048", "2304x1728", "1728x2304", "2496x1664", "1664x2496", "1792x2240", "2240x1792", "3136x1344", "2752x1536", "1536x2752",
|
||
"2880x2880", "3264x2448", "2448x3264", "3504x2336", "2336x3504", "2560x3200", "3200x2560", "3808x1632", "3840x2160", "2160x3840",
|
||
}
|
||
}
|
||
|
||
// openAIImageDefaultSizeTable 是 openai-image 默认能力配置的档位换算表（比例 × 1K/2K/4K）。
|
||
// 像素值与库内 openai-image 线路实际使用的预设逐项一致（gpt-image-2 / 2.5 / 4k 的 1K/2K/4K 全部命中），
|
||
// 且都在 defaultImageSizeValues() 里，不引入未声明的尺寸。
|
||
var openAIImageDefaultSizeTable = []struct {
|
||
ratio string
|
||
tierSizes [3]string
|
||
}{
|
||
{"1:1", [3]string{"1024x1024", "2048x2048", "2880x2880"}},
|
||
{"16:9", [3]string{"1824x1024", "2752x1536", "3840x2160"}},
|
||
{"9:16", [3]string{"1024x1824", "1536x2752", "2160x3840"}},
|
||
{"4:3", [3]string{"1360x1024", "2304x1728", "3264x2448"}},
|
||
{"3:4", [3]string{"1024x1360", "1728x2304", "2448x3264"}},
|
||
{"3:2", [3]string{"1536x1024", "2496x1664", "3504x2336"}},
|
||
{"2:3", [3]string{"1024x1536", "1664x2496", "2336x3504"}},
|
||
{"4:5", [3]string{"1024x1280", "1792x2240", "2560x3200"}},
|
||
{"5:4", [3]string{"1280x1024", "2240x1792", "3200x2560"}},
|
||
{"21:9", [3]string{"2048x878", "3136x1344", "3808x1632"}},
|
||
}
|
||
|
||
// openAIImageDefaultSizePresets 把上面的表展开成 3 个档位 × 10 个比例的预设。
|
||
// 像素值都来自 defaultImageSizeValues()，validateImageCapabilityConfig 会校验与支持值一致。
|
||
func openAIImageDefaultSizePresets() []ImageSizePreset {
|
||
tiers := []string{"1k", "2k", "4k"}
|
||
presets := make([]ImageSizePreset, 0, len(openAIImageDefaultSizeTable)*len(tiers))
|
||
for _, entry := range openAIImageDefaultSizeTable {
|
||
for index, tier := range tiers {
|
||
size := entry.tierSizes[index]
|
||
widthText, heightText, found := strings.Cut(size, "x")
|
||
width, widthErr := strconv.Atoi(widthText)
|
||
height, heightErr := strconv.Atoi(heightText)
|
||
if !found || widthErr != nil || heightErr != nil {
|
||
continue
|
||
}
|
||
presets = append(presets, ImageSizePreset{Tier: tier, Ratio: entry.ratio, Size: size, Width: width, Height: height})
|
||
}
|
||
}
|
||
return presets
|
||
}
|
||
|
||
// legacyImageSizeValues 用于修复旧数据中仅保存了 "*" 的图片尺寸能力。
|
||
// 这组值是前后台共同展示的基础预设，不能让历史通配符配置继续污染用户生成参数。
|
||
func legacyImageSizeValues() []string {
|
||
return []string{
|
||
"1:1", "3:2", "2:3", "4:3", "3:4", "16:9", "21:9", "9:16",
|
||
"1024x1024", "1536x1024", "1024x1536",
|
||
}
|
||
}
|
||
|
||
func DefaultModelCapabilityConfigForModel(protocol string, modelName string) *ModelCapabilityConfig {
|
||
// 文本模型是否支持视觉输入不能从协议或模型名可靠推断，默认关闭，由管理员按真实上游能力开启。
|
||
streaming := true
|
||
text := &TextCapabilityConfig{Streaming: &streaming, ContextWindowTokens: 128000, MaxOutputTokens: 16384, References: TextReferenceConfig{PromptMaxChars: 32000}}
|
||
video := &VideoCapabilityConfig{
|
||
References: VideoReferenceConfig{PromptMaxChars: DefaultVideoPromptMaxChars, MinImages: 0, MaxImages: 9, MaxImageBytes: 30 * 1024 * 1024, MaxVideos: 0, MaxVideoBytes: 0, MaxVideoDuration: 0, MaxAudios: 0, MaxAudioBytes: 0, MaxAudioDuration: 0},
|
||
Duration: VideoDurationConfig{Selection: "range", Min: 1, Max: 15, Step: 1, Default: 6},
|
||
Ratios: []string{"16:9", "9:16", "1:1", "4:3", "3:4", "21:9"},
|
||
DefaultRatio: "16:9",
|
||
Resolutions: []string{"480p", "720p", "1080p", "1440p", "2160p"},
|
||
DefaultResolution: "720p",
|
||
GenerateAudio: VideoBooleanConfig{Supported: false, Default: false},
|
||
Watermark: VideoBooleanConfig{Supported: false, Default: false},
|
||
Operations: []string{"text_to_video", "image_to_video"},
|
||
DefaultOperation: "text_to_video",
|
||
}
|
||
switch model.ChannelInterfaceType(protocol) {
|
||
case model.ChannelInterfaceVolcengineJiMengVideo:
|
||
video.Duration = VideoDurationConfig{Selection: "enum", Values: []int{5, 10}, Default: 5}
|
||
video.Resolutions = []string{"720p"}
|
||
case model.ChannelInterfaceGeminiVeo:
|
||
video.Duration = VideoDurationConfig{Selection: "enum", Values: []int{4, 6, 8}, Default: 6}
|
||
video.Resolutions = []string{"720p", "1080p"}
|
||
case model.ChannelInterfaceVolcengineArkVideo, model.ChannelInterfaceVolcengineArkAgentPlanVideo:
|
||
video.Operations = append(video.Operations, "reference_to_video", "audio_to_video")
|
||
video.References.MaxVideos, video.References.MaxAudios = 3, 3
|
||
video.References.MaxVideoBytes, video.References.MaxAudioBytes = 200*1024*1024, 15*1024*1024
|
||
video.References.MaxVideoDuration, video.References.MaxAudioDuration = 15, 15
|
||
video.GenerateAudio = VideoBooleanConfig{Supported: true, Default: true}
|
||
video.Watermark = VideoBooleanConfig{Supported: true, Default: false}
|
||
video.Resolutions = []string{"480p", "720p", "1080p"}
|
||
case model.ChannelInterfaceNewAPIChannel1, model.ChannelInterfaceNewAPIChannel2:
|
||
video.References.MaxVideos, video.References.MaxAudios = 3, 3
|
||
video.References.MaxVideoBytes, video.References.MaxAudioBytes = 200*1024*1024, 15*1024*1024
|
||
video.References.MaxVideoDuration, video.References.MaxAudioDuration = 15, 15
|
||
video.GenerateAudio = VideoBooleanConfig{Supported: true, Default: true}
|
||
if model.ChannelInterfaceType(protocol) == model.ChannelInterfaceNewAPIChannel1 {
|
||
video.Resolutions = []string{"480p", "720p", "1080p"}
|
||
}
|
||
case model.ChannelInterfaceNewAPIVideo, model.ChannelInterfaceXAIVideo:
|
||
video.GenerateAudio = VideoBooleanConfig{Supported: false, Default: false}
|
||
case model.ChannelInterfaceNovitaVideo:
|
||
video.References.MaxImages, video.References.MaxImageBytes = 1, 10*1024*1024
|
||
video.Duration = VideoDurationConfig{Selection: "enum", Values: []int{5, 10}, Default: 5}
|
||
video.Ratios = []string{"16:9", "9:16", "1:1"}
|
||
video.Resolutions = []string{"1080p"}
|
||
video.DefaultResolution = "1080p"
|
||
case model.ChannelInterfaceMiniMaxVideo:
|
||
video.Operations = append(video.Operations, "reference_to_video")
|
||
video.References.MaxImages = 9
|
||
video.References.MaxImageBytes = 30 * 1024 * 1024
|
||
video.References.MaxVideos = 3
|
||
video.References.MaxVideoBytes = 50 * 1024 * 1024
|
||
video.References.MaxVideoDuration = 15
|
||
video.References.MaxAudios = 3
|
||
video.References.MaxAudioBytes = 15 * 1024 * 1024
|
||
video.References.MaxAudioDuration = 15
|
||
video.Duration = VideoDurationConfig{Selection: "enum", Values: []int{4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}, Default: 5}
|
||
video.Ratios = []string{"adaptive", "21:9", "16:9", "4:3", "1:1", "3:4", "9:16"}
|
||
video.DefaultRatio = "16:9"
|
||
video.Resolutions = []string{"768P", "2K"}
|
||
video.DefaultResolution = "768P"
|
||
video.Watermark = VideoBooleanConfig{Supported: true, Default: false}
|
||
case model.ChannelInterfaceAgnesVideo:
|
||
video = applyModelSpecificVideoCapability(video, protocol, modelName)
|
||
}
|
||
return &ModelCapabilityConfig{Version: 1, Text: text, Image: DefaultImageCapabilityConfig(protocol, modelName), Video: video}
|
||
}
|