Files
image2api/backend/internal/provider/adobe/payloads.go
T

518 lines
19 KiB
Go

package adobe
import (
"encoding/json"
"math/rand"
"strings"
)
type modelSpec struct {
UpstreamModelID string
UpstreamModelVersion string
}
var lumaSize = map[string]map[string][2]int{
"720p": {
"21:9": {1280, 548}, "16:9": {1280, 720}, "4:3": {960, 720},
"1:1": {720, 720}, "3:4": {720, 960}, "9:16": {720, 1280}, "9:21": {548, 1280},
},
"1080p": {
"21:9": {1920, 822}, "16:9": {1920, 1080}, "4:3": {1440, 1080},
"1:1": {1080, 1080}, "3:4": {1080, 1440}, "9:16": {1080, 1920}, "9:21": {822, 1920},
},
"4k": {
"21:9": {3840, 1646}, "16:9": {3840, 2160}, "4:3": {2880, 2160},
"1:1": {2160, 2160}, "3:4": {2160, 2880}, "9:16": {2160, 3840}, "9:21": {1646, 3840},
},
}
var gptImageSize = map[string]map[string][2]int{
"1K": {"1:1": {1024, 1024}, "5:4": {1120, 896}, "9:16": {720, 1280}, "21:9": {1456, 624}, "16:9": {1280, 720}, "4:3": {1152, 864}, "3:2": {1248, 832}, "4:5": {896, 1120}, "3:4": {864, 1152}, "2:3": {832, 1248}},
"2K": {"1:1": {2048, 2048}, "5:4": {2240, 1792}, "9:16": {1440, 2560}, "21:9": {3024, 1296}, "16:9": {2560, 1440}, "4:3": {2304, 1728}, "3:2": {2496, 1664}, "4:5": {1792, 2240}, "3:4": {1728, 2304}, "2:3": {1664, 2496}},
"4K": {"1:1": {2880, 2880}, "5:4": {3200, 2560}, "9:16": {2160, 3840}, "21:9": {3696, 1584}, "16:9": {3840, 2160}, "4:3": {3264, 2448}, "3:2": {3504, 2336}, "4:5": {2560, 3200}, "3:4": {2448, 3264}, "2:3": {2336, 3504}},
}
var fluxSize = map[string][2]int{
"1:1": {1024, 1024},
"16:9": {1408, 768},
"9:16": {768, 1408},
"4:3": {1280, 896},
"3:4": {896, 1280},
}
var defaultSize = map[string]map[string][2]int{
"1K": {"1:1": {1024, 1024}, "1:8": {384, 3072}, "1:4": {512, 2048}, "16:9": {1360, 768}, "9:16": {768, 1360}, "4:1": {2048, 512}, "4:3": {1152, 864}, "3:4": {864, 1152}, "8:1": {3072, 384}},
"2K": {"1:1": {2048, 2048}, "1:8": {768, 6144}, "1:4": {1024, 4096}, "16:9": {2752, 1536}, "9:16": {1536, 2752}, "4:1": {4096, 1024}, "4:3": {2048, 1536}, "3:4": {1536, 2048}, "8:1": {6144, 768}},
"4K": {"1:1": {4096, 4096}, "1:8": {1536, 12288}, "1:4": {2048, 8192}, "16:9": {5504, 3072}, "9:16": {3072, 5504}, "4:1": {8192, 2048}, "4:3": {4096, 3072}, "3:4": {3072, 4096}, "8:1": {12288, 1536}},
}
func ResolveModelSpec(modelID string) modelSpec {
switch modelID {
case "firefly-gpt-image", "firefly-gpt-image-2":
return modelSpec{UpstreamModelID: "gpt-image", UpstreamModelVersion: "2"}
case "flux-kontext-max":
return modelSpec{UpstreamModelID: "flux", UpstreamModelVersion: "fluxKontextMax"}
case "nano-banana-2":
return modelSpec{UpstreamModelID: "gemini-flash", UpstreamModelVersion: "nano-banana-3"}
case "nano-banana-pro":
return modelSpec{UpstreamModelID: "gemini-flash", UpstreamModelVersion: "nano-banana-2"}
default:
return modelSpec{UpstreamModelID: "gemini-flash", UpstreamModelVersion: "gemini-3-pro-image-preview"}
}
}
// buildImage5Payload builds the Adobe Firefly Image 5 request. It uses a distinct
// schema from the firefly-3p models: NO modelId/size, a top-level aspectRatio
// string label and a resolutionLevel (1K→1MP, 2K→4MP). Mirrors a captured
// working image-v5.ff.adobe.io request.
func buildImage5Payload(prompt, aspectRatio, resolution string, blobIDs []string) map[string]any {
p := map[string]any{
"n": 1,
"seeds": []int{rand.Intn(999999)},
"output": map[string]any{"storeInputs": true},
"prompt": prompt,
"referenceBlobs": []any{},
"modelSpecificPayload": map[string]any{"locale": "en-US", "prompt_reasoner": "quality"},
"modelVersion": "image5",
"resolutionLevel": image5ResolutionLevel(resolution),
"generationMetadata": map[string]any{"module": "text2image", "submodule": "ff-image-generate"},
}
if len(blobIDs) > 0 {
// Instruct-edit: aspect ratio is derived from the reference image; sending
// aspectRatio is rejected with a validation_error.
p["referenceBlobs"] = blobRefs(blobIDs, "general")
} else {
p["aspectRatio"] = defaultString(aspectRatio, "1:1")
}
return p
}
// image5ResolutionLevel maps the UI resolution tier to Image 5's megapixel level.
func image5ResolutionLevel(resolution string) string {
switch strings.ToUpper(strings.TrimSpace(resolution)) {
case "1K":
return "1MP"
case "2K":
return "4MP"
default:
return "4MP"
}
}
func BuildImagePayloadCandidates(modelID, prompt, aspectRatio, outputResolution string, blobIDs []string) []map[string]any {
spec := ResolveModelSpec(modelID)
ratio := defaultString(aspectRatio, "1:1")
resolution := defaultString(outputResolution, "2K")
switch spec.UpstreamModelID {
case "gpt-image":
return buildGPTImagePayloads(spec, prompt, ratio, resolution, blobIDs)
case "flux":
return buildFluxPayloads(spec, prompt, ratio, blobIDs)
default:
return buildDefaultPayloads(spec, prompt, ratio, resolution, blobIDs)
}
}
func buildGPTImagePayloads(spec modelSpec, prompt, ratio, resolution string, blobIDs []string) []map[string]any {
size := getSize(gptImageSize, resolution, ratio, "1:1")
// Mirrors the captured working gpt-image request shape: modelSpecificPayload.size,
// generationSettings.detailLevel 3, and NO top-level size / outputResolution
// (sending those got 403). Keeps the chosen size via modelSpecificPayload.size
// ("WxH") rather than "auto".
base := map[string]any{
"modelId": spec.UpstreamModelID,
"modelVersion": spec.UpstreamModelVersion,
"n": 1,
"prompt": prompt,
"seeds": []int{rand.Intn(999999)},
"output": map[string]any{"storeInputs": true},
"referenceBlobs": []any{},
"generationMetadata": map[string]any{"module": "text2image", "submodule": "ff-image-generate"},
"modelSpecificPayload": map[string]any{"size": sizeString(size)},
"generationSettings": map[string]any{"detailLevel": 3},
}
if len(blobIDs) == 0 {
return []map[string]any{base}
}
subject := cloneMap(base)
subject["referenceBlobs"] = blobRefs(blobIDs, "subject")
return []map[string]any{subject}
}
func buildFluxPayloads(spec modelSpec, prompt, ratio string, blobIDs []string) []map[string]any {
size := fluxSize[ratio]
if size == [2]int{} {
size = fluxSize["1:1"]
}
base := map[string]any{
"modelId": spec.UpstreamModelID,
"modelVersion": spec.UpstreamModelVersion,
"n": 1,
"prompt": prompt,
"size": map[string]any{"width": size[0], "height": size[1]},
"seeds": []int{rand.Intn(999999)},
"output": map[string]any{"storeInputs": true},
"referenceBlobs": []any{},
"modelSpecificPayload": map[string]any{
"prompt_upsampling": true,
"safety_tolerance": 2,
"aspect_ratio": ratio,
},
"generationMetadata": map[string]any{"module": "text2image", "submodule": "ff-image-generate"},
}
if len(blobIDs) == 0 {
return []map[string]any{base}
}
edited := cloneMap(base)
edited["generationMetadata"] = map[string]any{"module": "image2image", "submodule": "ff-image-generate"}
edited["referenceBlobs"] = blobRefs(blobIDs, "general")
return []map[string]any{edited}
}
func buildDefaultPayloads(spec modelSpec, prompt, ratio, resolution string, blobIDs []string) []map[string]any {
size := getSize(defaultSize, resolution, ratio, "16:9")
// Shape mirrors a captured working firefly.adobe.com request exactly: top-level
// size object, modelSpecificPayload only {parameters:{addWatermark:false}},
// groundSearch:false, module "text2image" (even with a reference blob). NO
// skipCai and NO modelSpecificPayload.aspectRatio — sending those got 403.
base := map[string]any{
"modelId": spec.UpstreamModelID,
"modelVersion": spec.UpstreamModelVersion,
"n": 1,
"prompt": prompt,
"size": map[string]any{"width": size[0], "height": size[1]},
"seeds": []int{rand.Intn(999999)},
"groundSearch": false,
"output": map[string]any{"storeInputs": true},
"generationMetadata": map[string]any{
"module": "text2image",
"submodule": "ff-image-generate",
},
"modelSpecificPayload": map[string]any{
"parameters": map[string]any{"addWatermark": false},
},
}
if len(blobIDs) == 0 {
base["referenceBlobs"] = []any{}
return []map[string]any{base}
}
edited := cloneMap(base)
edited["referenceBlobs"] = blobRefs(blobIDs, "general")
return []map[string]any{edited}
}
func getSize(table map[string]map[string][2]int, resolution, ratio, fallbackRatio string) [2]int {
level := defaultString(resolution, "2K")
levelTable, ok := table[level]
if !ok {
levelTable = table["2K"]
}
size, ok := levelTable[ratio]
if !ok {
size = levelTable[fallbackRatio]
}
return size
}
func sizeString(size [2]int) string {
return itoa(size[0]) + "x" + itoa(size[1])
}
func blobRefs(ids []string, usage string) []any {
out := make([]any, 0, len(ids))
for _, id := range ids {
out = append(out, map[string]any{"id": id, "usage": usage})
}
return out
}
func referenceImagesByID(ids []string) []any {
out := make([]any, 0, len(ids))
for _, id := range ids {
out = append(out, map[string]any{"id": id})
}
return out
}
func referenceImagesByLocal(ids []string) []any {
out := make([]any, 0, len(ids))
for _, id := range ids {
out = append(out, map[string]any{"localBlobRef": id})
}
return out
}
func cloneMap(in map[string]any) map[string]any {
out := make(map[string]any, len(in))
for k, v := range in {
out[k] = v
}
return out
}
func BuildVideoPayload(engine, prompt, aspectRatio string, durationSeconds int, resolution, referenceMode, upstreamModel string, blobIDs, videoBlobIDs, audioBlobIDs []string) map[string]any {
seedVal := rand.Intn(999999)
engine = defaultString(engine, "sora2")
resolution = defaultString(resolution, "720p")
aspectRatio = defaultString(aspectRatio, "16:9")
if durationSeconds <= 0 {
durationSeconds = 5
}
switch engine {
case "firefly-video":
// Firefly-native video model — a distinct schema (mirrors a captured
// working video-v1.ff.adobe.io request): sizes[] carries width/height +
// numFrames (numFrames encodes duration, ~25.6fps so 5s = 128), and
// reference frames go under image.conditions with placement.start
// (0 = first frame / 首帧, 1 = last frame / 末帧). NO modelId / version /
// engine / duration / referenceBlobs fields.
w, h, frames := fireflyVideoSize(aspectRatio, resolution, durationSeconds)
payload := map[string]any{
"addOnTransparentBackground": false,
"prompt": prompt,
"seeds": []int{seedVal},
"sizes": []any{map[string]any{"width": w, "height": h, "numFrames": frames}},
"videoSettings": map[string]any{},
"locale": "en-US",
"generationMetadata": map[string]any{"module": "text2video", "submodule": "ff-video-generate"},
"output": map[string]any{"storeInputs": true},
}
if len(blobIDs) > 0 {
conds := make([]any, 0, 2)
conds = append(conds, map[string]any{
"source": map[string]any{"id": blobIDs[0]},
"placement": map[string]any{"start": 0},
})
if len(blobIDs) > 1 {
conds = append(conds, map[string]any{
"source": map[string]any{"id": blobIDs[1]},
"placement": map[string]any{"start": 1},
})
}
payload["image"] = map[string]any{"conditions": conds}
}
return payload
case "veo31-fast", "veo31-standard":
modelVersion := "3.1-fast-generate"
if engine == "veo31-standard" {
modelVersion = "3.1-generate"
}
payload := map[string]any{
"modelId": "veo",
"modelVersion": modelVersion,
"size": videoSize(aspectRatio, resolution),
"seeds": []int{seedVal},
"prompt": prompt,
"negativePrompt": "",
"duration": durationSeconds,
"generateAudio": false,
"generationMetadata": map[string]any{
"module": "text2video",
"submodule": "ff-video-generate",
},
"output": map[string]any{"storeInputs": true},
"referenceBlobs": []any{},
}
if len(blobIDs) > 0 {
payload["generationMetadata"] = map[string]any{"module": "image2video", "submodule": "ff-video-generate"}
refs := make([]any, 0, min(len(blobIDs), 3))
if referenceMode == "frame" {
for idx, id := range blobIDs[:min(len(blobIDs), 3)] {
refs = append(refs, map[string]any{"id": id, "usage": "frame", "order": idx + 1})
}
} else {
for _, id := range blobIDs[:min(len(blobIDs), 3)] {
refs = append(refs, map[string]any{"id": id, "usage": "asset"})
}
}
payload["referenceBlobs"] = refs
}
return payload
case "seedance-2.0-fast", "seedance-2.0":
modelVersion := "seedance_2.0"
if engine == "seedance-2.0-fast" {
modelVersion = "seedance_2.0_fast"
}
payload := map[string]any{
"modelId": "seedance",
"modelVersion": modelVersion,
"size": videoSize(aspectRatio, resolution),
"seeds": []int{seedVal},
"prompt": prompt,
"negativePrompt": "",
"duration": durationSeconds,
"generateAudio": true,
"generationSettings": map[string]any{"aspectRatio": aspectRatio},
"generationMetadata": map[string]any{"module": "text2video", "submodule": "ff-video-generate"},
"output": map[string]any{"storeInputs": true},
"referenceBlobs": []any{},
}
if len(blobIDs) > 0 {
// Seedance keeps module "text2video" even with reference blobs
// (HAR-confirmed) — unlike veo/luma it does not switch to image2video.
if referenceMode == "frame" {
refs := make([]any, 0, min(len(blobIDs), 2))
for idx, id := range blobIDs[:min(len(blobIDs), 2)] {
refs = append(refs, map[string]any{"id": id, "usage": "frame", "order": idx + 1})
}
payload["referenceBlobs"] = refs
} else {
refs := make([]any, 0, min(len(blobIDs), 9))
for _, id := range blobIDs[:min(len(blobIDs), 9)] {
refs = append(refs, map[string]any{"id": id, "usage": "style"})
}
payload["referenceBlobs"] = refs
}
}
// Video/audio source refs (seedance)
for _, id := range videoBlobIDs[:min(len(videoBlobIDs), 3)] {
payload["referenceBlobs"] = append(payload["referenceBlobs"].([]any), map[string]any{"id": id, "usage": "source"})
}
for _, id := range audioBlobIDs[:min(len(audioBlobIDs), 3)] {
payload["referenceBlobs"] = append(payload["referenceBlobs"].([]any), map[string]any{"id": id, "usage": "source"})
}
return payload
case "luma":
payload := map[string]any{
"modelId": "luma",
"modelVersion": "3.14-ray",
"size": lumaVideoSize(aspectRatio, resolution),
"mode": "flex_2",
"prompt": prompt,
"negativePrompt": "",
"duration": durationSeconds,
"generationMetadata": map[string]any{
"module": "text2video",
"submodule": "ff-video-generate",
},
"modelSpecificPayload": map[string]any{
"resolution": strings.ToLower(resolution),
"aspect_ratio": aspectRatio,
},
"output": map[string]any{"storeInputs": true},
}
if len(blobIDs) > 0 {
payload["generationMetadata"] = map[string]any{
"module": "image2video",
"submodule": "ff-video-generate",
}
refs := make([]any, 0, min(len(blobIDs), 2))
for idx, id := range blobIDs[:min(len(blobIDs), 2)] {
refs = append(refs, map[string]any{"id": id, "usage": "frame", "order": idx + 1})
}
payload["referenceBlobs"] = refs
}
return payload
default:
upstream := defaultString(upstreamModel, "openai:firefly:colligo:sora2")
payload := map[string]any{
"n": 1,
"seeds": []int{seedVal},
"modelId": "sora",
"modelVersion": "sora-2",
"size": videoSize(aspectRatio, resolution),
"duration": durationSeconds,
"fps": 24,
"prompt": buildVideoPromptJSON(prompt, durationSeconds),
"generationMetadata": map[string]any{"module": "text2video"},
"model": upstream,
"generateAudio": true,
"generateLoop": false,
"transparentBackground": false,
"seed": itoa(seedVal),
"locale": "en-US",
"camera": map[string]any{"angle": "none", "shotSize": "none", "motion": nil, "promptStyle": nil},
"negativePrompt": "",
"jobMode": "standard",
"debugGenerationEndpoint": "",
"referenceBlobs": []any{},
"referenceFrames": []any{},
"referenceVideo": nil,
"cameraMotionReferenceVideo": nil,
"characterReference": nil,
"editReferenceVideo": nil,
"output": map[string]any{"storeInputs": true},
}
if len(blobIDs) > 0 {
firstID := blobIDs[0]
payload["generationMetadata"] = map[string]any{"module": "image2video"}
payload["referenceBlobs"] = []any{
map[string]any{"id": firstID, "usage": "general", "promptReference": 1},
}
payload["referenceFrames"] = []any{map[string]any{"localBlobRef": firstID}, nil}
}
return payload
}
}
// fireflyVideoSizeTable maps the firefly-video resolution tier + aspect ratio to
// pixel dimensions. Only 1080p 9:16 (1080x1920) is HAR-confirmed; the rest follow
// the standard 540p/720p/1080p grid for each ratio.
var fireflyVideoSizeTable = map[string]map[string][2]int{
"540p": {"16:9": {960, 540}, "1:1": {540, 540}, "9:16": {540, 960}},
"720p": {"16:9": {1280, 720}, "1:1": {720, 720}, "9:16": {720, 1280}},
"1080p": {"16:9": {1920, 1080}, "1:1": {1080, 1080}, "9:16": {1080, 1920}},
}
// fireflyVideoSize returns width, height and numFrames. numFrames encodes the
// clip length (~25.6fps; 5s = 128 frames, HAR-confirmed).
func fireflyVideoSize(aspectRatio, resolution string, durationSeconds int) (int, int, int) {
table, ok := fireflyVideoSizeTable[strings.ToLower(defaultString(resolution, "1080p"))]
if !ok {
table = fireflyVideoSizeTable["1080p"]
}
wh, ok := table[defaultString(aspectRatio, "9:16")]
if !ok {
wh = table["9:16"]
}
frames := durationSeconds * 128 / 5
if frames <= 0 {
frames = 128
}
return wh[0], wh[1], frames
}
func videoSize(aspectRatio, resolution string) map[string]any {
if strings.EqualFold(resolution, "1080p") {
if aspectRatio == "16:9" {
return map[string]any{"width": 1920, "height": 1080}
}
return map[string]any{"width": 1080, "height": 1920}
}
if aspectRatio == "16:9" {
return map[string]any{"width": 1280, "height": 720}
}
return map[string]any{"width": 720, "height": 1280}
}
func lumaVideoSize(aspectRatio, resolution string) map[string]any {
table, ok := lumaSize[strings.ToLower(defaultString(resolution, "720p"))]
if !ok {
table = lumaSize["720p"]
}
size, ok := table[defaultString(aspectRatio, "16:9")]
if !ok {
size = table["16:9"]
}
return map[string]any{"width": size[0], "height": size[1]}
}
func buildVideoPromptJSON(prompt string, durationSeconds int) string {
payload := map[string]any{
"id": 1,
"duration_sec": durationSeconds,
"prompt_text": prompt,
}
b, _ := json.Marshal(payload)
return string(b)
}
func min(a, b int) int {
if a < b {
return a
}
return b
}