mirror of
https://github.com/HeyPuter/puter.git
synced 2026-05-04 08:30:39 +00:00
fix gemini image provider implementation (#2569)
* implement proper calls to gemini image * remove hardcoded image prices * fix model selection * add flash 3.1 image, fix pricing * also handle uris with data prefix * more efficient uri parsing * fix token calculation * re-add some costs for models page
This commit is contained in:
Shruc
@@ -125,6 +125,7 @@ class AIInterfaceService extends BaseService {
|
||||
prompt: { type: 'string' },
|
||||
quality: { type: 'string' },
|
||||
model: { type: 'string' },
|
||||
provider: { type: 'string', optional: true },
|
||||
ratio: { type: 'json' },
|
||||
width: { type: 'number', optional: true },
|
||||
height: { type: 'number', optional: true },
|
||||
@@ -135,6 +136,7 @@ class AIInterfaceService extends BaseService {
|
||||
n: { type: 'number', optional: true },
|
||||
input_image: { type: 'string', optional: true },
|
||||
input_image_mime_type: { type: 'string', optional: true },
|
||||
input_images: { type: 'json', optional: true },
|
||||
image_url: { type: 'string', optional: true },
|
||||
image_base64: { type: 'string', optional: true },
|
||||
mask_image_url: { type: 'string', optional: true },
|
||||
|
||||
@@ -68,13 +68,13 @@ export class AIImageGenerationService extends BaseService {
|
||||
|
||||
/** Driver interfaces */
|
||||
static IMPLEMENTS = {
|
||||
['driver-capabilities']: {
|
||||
'driver-capabilities': {
|
||||
supports_test_mode (iface: string, method_name: string) {
|
||||
return iface === 'puter-image-generation' &&
|
||||
method_name === 'generate';
|
||||
},
|
||||
},
|
||||
['puter-image-generation']: {
|
||||
'puter-image-generation': {
|
||||
|
||||
async generate (...parameters: Parameters<AIImageGenerationService['generate']>) {
|
||||
return (this as unknown as AIImageGenerationService).generate(...parameters);
|
||||
@@ -88,11 +88,18 @@ export class AIImageGenerationService extends BaseService {
|
||||
return undefined;
|
||||
}
|
||||
|
||||
if ( ! provider ) {
|
||||
return models[0];
|
||||
if ( provider ) {
|
||||
const model = models.find(m => m.provider === provider);
|
||||
return model ?? models[0];
|
||||
}
|
||||
const model = models.find(m => m.provider === provider);
|
||||
return model ?? models[0];
|
||||
|
||||
// If no provider is specified, prefer a model whose puterId exactly matches the requested modelId.
|
||||
const exactPuterIdMatch = models.find(m => m.puterId === modelId);
|
||||
if ( exactPuterIdMatch ) {
|
||||
return exactPuterIdMatch;
|
||||
}
|
||||
|
||||
return models[0];
|
||||
}
|
||||
|
||||
private async registerProviders () {
|
||||
@@ -152,9 +159,11 @@ export class AIImageGenerationService extends BaseService {
|
||||
const provider = this.#providers[providerName];
|
||||
|
||||
// alias all driver requests to go here to support legacy routing
|
||||
this.driverService.register_service_alias(AIImageGenerationService.SERVICE_NAME,
|
||||
providerName,
|
||||
{ iface: 'puter-image-generation' });
|
||||
this.driverService.register_service_alias(
|
||||
AIImageGenerationService.SERVICE_NAME,
|
||||
providerName,
|
||||
{ iface: 'puter-image-generation' },
|
||||
);
|
||||
|
||||
// build model id map
|
||||
for ( const model of await provider.models() ) {
|
||||
|
||||
+139
-56
@@ -22,17 +22,20 @@ import APIError from '../../../../../api/APIError.js';
|
||||
import { ErrorService } from '../../../../../modules/core/ErrorService.js';
|
||||
import { Context } from '../../../../../util/context.js';
|
||||
import { MeteringService } from '../../../../MeteringService/MeteringService.js';
|
||||
import { GEMINI_DEFAULT_RATIO, GEMINI_IMAGE_GENERATION_MODELS } from './models.js';
|
||||
import { GEMINI_DEFAULT_RATIO, GEMINI_ESTIMATED_IMAGE_TOKENS, GEMINI_IMAGE_GENERATION_MODELS } from './models.js';
|
||||
import { IGenerateParams, IImageModel, IImageProvider } from '../types.js';
|
||||
|
||||
type GeminiGenerateParams = IGenerateParams & {
|
||||
input_image?: string;
|
||||
input_image_mime_type?: string;
|
||||
const MIME_SIGNATURES: Record<string, string> = {
|
||||
'/9j/': 'image/jpeg',
|
||||
'iVBOR': 'image/png',
|
||||
'UklGR': 'image/webp',
|
||||
};
|
||||
|
||||
interface GeminiUsageMetadata {
|
||||
promptTokenCount: number;
|
||||
candidatesTokenCount: number;
|
||||
candidatesTextTokenCount: number;
|
||||
candidatesImageTokenCount: number;
|
||||
thoughtsTokenCount: number;
|
||||
}
|
||||
|
||||
@@ -59,9 +62,8 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
}
|
||||
|
||||
async generate (params: IGenerateParams): Promise<string> {
|
||||
const { prompt, test_mode } = params;
|
||||
let { model, ratio, quality } = params;
|
||||
const { input_image, input_image_mime_type } = params as GeminiGenerateParams;
|
||||
const { prompt, test_mode, input_image, input_image_mime_type, model, quality } = params;
|
||||
let { ratio, input_images } = params;
|
||||
|
||||
const selectedModel = this.models().find(m => m.id === model) || this.models().find(m => m.id === this.getDefaultModel())!;
|
||||
|
||||
@@ -76,28 +78,19 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
const allowedRatios = selectedModel.allowedRatios ?? [GEMINI_DEFAULT_RATIO];
|
||||
ratio = ratio && this.#isValidRatio(ratio, allowedRatios) ? ratio : allowedRatios[0];
|
||||
|
||||
if ( input_image && !input_image_mime_type ) {
|
||||
throw new Error('`input_image_mime_type` is required when `input_image` is provided');
|
||||
// Backwards compat: merge singular input_image into input_images
|
||||
if ( input_image && (!input_images || input_images.length === 0) ) {
|
||||
input_images = [input_image];
|
||||
}
|
||||
|
||||
if ( input_image_mime_type && !input_image ) {
|
||||
throw new Error('`input_image` is required when `input_image_mime_type` is provided');
|
||||
}
|
||||
|
||||
if ( input_image_mime_type && !this.#isValidImageMimeType(input_image_mime_type) ) {
|
||||
throw new Error('`input_image_mime_type` must be a valid image MIME type (image/png, image/jpeg, image/webp)');
|
||||
}
|
||||
|
||||
const priceKey = `${quality ? `${quality}:` : ''}${ratio.w}x${ratio.h}`;
|
||||
const priceInCents = selectedModel.costs[priceKey];
|
||||
if ( priceInCents === undefined ) {
|
||||
const availableSizes = Object.keys(selectedModel.costs)
|
||||
.filter(key => key !== 'input' && key !== 'output');
|
||||
throw APIError.create('field_invalid', undefined, {
|
||||
key: 'size/quality combination',
|
||||
expected: `one of: ${ availableSizes.join(', ')}`,
|
||||
got: priceKey,
|
||||
});
|
||||
// Validate input images have detectable MIME types
|
||||
if ( input_images?.length ) {
|
||||
for ( const img of input_images ) {
|
||||
const mime = this.#detectMimeType(img) ?? input_image_mime_type;
|
||||
if ( ! mime ) {
|
||||
throw new Error('Could not detect MIME type for an input image. Provide a known image format (JPEG, PNG, WebP) or set `input_image_mime_type`.');
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
const actor = Context.get('actor');
|
||||
@@ -110,9 +103,22 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
});
|
||||
}
|
||||
|
||||
const estimatedPromptTokenCount = this.#estimatePromptTokenCount(prompt);
|
||||
// --- Pre-flight cost estimation ---
|
||||
const inputImageCount = input_images?.length ?? 0;
|
||||
const estimatedImageInputTokens = inputImageCount * 560; // https://ai.google.dev/gemini-api/docs/pricing#gemini-3-pro-image-preview
|
||||
const estimatedPromptTokenCount = this.#estimatePromptTokenCount(prompt) + estimatedImageInputTokens;
|
||||
const estimatedInputCostInCents = this.#calculateTokenCostInCents(estimatedPromptTokenCount, selectedModel.costs.input);
|
||||
const estimatedOutputCostInCents = priceInCents;
|
||||
|
||||
// Estimate output image tokens
|
||||
const imageTokenKey = quality ? `${selectedModel.id}:${quality}` : selectedModel.id;
|
||||
const estimatedOutputImageTokens = GEMINI_ESTIMATED_IMAGE_TOKENS[imageTokenKey] ?? GEMINI_ESTIMATED_IMAGE_TOKENS[selectedModel.id];
|
||||
if ( estimatedOutputImageTokens === undefined ) {
|
||||
throw new Error(`No estimated image token count configured for '${imageTokenKey}'.`);
|
||||
}
|
||||
const estimatedOutputImageCostInCents = this.#calculateTokenCostInCents(estimatedOutputImageTokens, selectedModel.costs.output_image);
|
||||
const estimatedOutputTextCostInCents = this.#calculateTokenCostInCents(50, selectedModel.costs.output); // small text overhead estimate
|
||||
const estimatedOutputCostInCents = estimatedOutputImageCostInCents + estimatedOutputTextCostInCents;
|
||||
|
||||
const estimatedTotalCostInMicroCents = this.#toMicroCents(estimatedInputCostInCents + estimatedOutputCostInCents);
|
||||
const usageAllowed = await this.#meteringService.hasEnoughCredits(actor, estimatedTotalCostInMicroCents);
|
||||
|
||||
@@ -120,21 +126,38 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
throw APIError.create('insufficient_funds');
|
||||
}
|
||||
|
||||
const contents = this.#buildContents(prompt, ratio, input_image, input_image_mime_type);
|
||||
// --- API call ---
|
||||
const contents = this.#buildContents(prompt, input_images, input_image_mime_type);
|
||||
const aspectRatio = `${ratio.w}:${ratio.h}`;
|
||||
|
||||
const imageConfig: Record<string, string> = { aspectRatio };
|
||||
if ( quality && selectedModel.allowedQualityLevels?.includes(quality) ) {
|
||||
imageConfig.imageSize = quality;
|
||||
}
|
||||
|
||||
const response = await this.#client.models.generateContent({
|
||||
model: selectedModel.id,
|
||||
contents,
|
||||
config: {
|
||||
responseModalities: ['TEXT', 'IMAGE'],
|
||||
imageConfig,
|
||||
},
|
||||
});
|
||||
|
||||
// --- Actual cost calculation from response usage ---
|
||||
const usage = this.#extractUsageMetadata(response);
|
||||
const inputTokenCount = usage.promptTokenCount || estimatedPromptTokenCount;
|
||||
const outputTokenCount = usage.candidatesTokenCount + usage.thoughtsTokenCount;
|
||||
|
||||
const outputTextTokenCount = usage.candidatesTextTokenCount + usage.thoughtsTokenCount;
|
||||
const outputImageTokenCount = usage.candidatesImageTokenCount || estimatedOutputImageTokens;
|
||||
|
||||
const inputCostInCents = this.#calculateTokenCostInCents(inputTokenCount, selectedModel.costs.input);
|
||||
const outputTextCostInCents = this.#calculateTokenCostInCents(outputTokenCount, selectedModel.costs.output);
|
||||
const outputCostInCents = priceInCents + outputTextCostInCents;
|
||||
const outputTextCostInCents = this.#calculateTokenCostInCents(outputTextTokenCount, selectedModel.costs.output);
|
||||
const outputImageCostInCents = this.#calculateTokenCostInCents(outputImageTokenCount, selectedModel.costs.output_image);
|
||||
const outputCostInCents = outputTextCostInCents + outputImageCostInCents;
|
||||
|
||||
const usagePrefix = `gemini:${selectedModel.id}:${priceKey}`;
|
||||
const totalOutputTokenCount = outputTextTokenCount + outputImageTokenCount;
|
||||
const usagePrefix = `gemini:${selectedModel.id}`;
|
||||
this.#meteringService.batchIncrementUsages(actor, [
|
||||
{
|
||||
usageType: `${usagePrefix}:input`,
|
||||
@@ -142,9 +165,14 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
costOverride: this.#toMicroCents(inputCostInCents),
|
||||
},
|
||||
{
|
||||
usageType: `${usagePrefix}:output`,
|
||||
usageAmount: Math.max(outputTokenCount, 1),
|
||||
costOverride: this.#toMicroCents(outputCostInCents),
|
||||
usageType: `${usagePrefix}:output:text`,
|
||||
usageAmount: Math.max(outputTextTokenCount, 1),
|
||||
costOverride: this.#toMicroCents(outputTextCostInCents),
|
||||
},
|
||||
{
|
||||
usageType: `${usagePrefix}:output:image`,
|
||||
usageAmount: Math.max(outputImageTokenCount, 1),
|
||||
costOverride: this.#toMicroCents(outputImageCostInCents),
|
||||
},
|
||||
]);
|
||||
|
||||
@@ -155,7 +183,9 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
inputCostInCents,
|
||||
outputCostInCents,
|
||||
inputTokenCount,
|
||||
outputTokenCount,
|
||||
outputTokenCount: totalOutputTokenCount,
|
||||
outputTextTokenCount,
|
||||
outputImageTokenCount,
|
||||
});
|
||||
|
||||
const url = this.#extractImageUrl(response);
|
||||
@@ -167,20 +197,24 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
return url;
|
||||
}
|
||||
|
||||
#buildContents (prompt: string, ratio: { w: number; h: number }, input_image?: string, input_image_mime_type?: string) {
|
||||
if ( input_image && input_image_mime_type ) {
|
||||
return [
|
||||
{ text: `Generate a picture of dimensions ${parseInt(`${ratio.w}`)}x${parseInt(`${ratio.h}`)} with the prompt: ${prompt}` },
|
||||
{
|
||||
#buildContents (prompt: string, input_images?: string[], input_image_mime_type?: string) {
|
||||
const parts: Record<string, unknown>[] = [{ text: prompt }];
|
||||
|
||||
if ( input_images?.length ) {
|
||||
for ( const img of input_images ) {
|
||||
const parsed = this.#parseDataUri(img);
|
||||
const mimeType = parsed?.mimeType ?? this.#detectMimeType(img) ?? input_image_mime_type ?? 'image/png';
|
||||
const rawBase64 = parsed?.base64 ?? img;
|
||||
parts.push({
|
||||
inlineData: {
|
||||
mimeType: input_image_mime_type,
|
||||
data: input_image,
|
||||
mimeType,
|
||||
data: rawBase64,
|
||||
},
|
||||
},
|
||||
];
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
return `Generate a picture of dimensions ${parseInt(`${ratio.w}`)}x${parseInt(`${ratio.h}`)} with the prompt: ${prompt}`;
|
||||
return parts;
|
||||
}
|
||||
|
||||
#setResponseCostMetadata ({
|
||||
@@ -191,6 +225,8 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
outputCostInCents,
|
||||
inputTokenCount,
|
||||
outputTokenCount,
|
||||
outputTextTokenCount,
|
||||
outputImageTokenCount,
|
||||
}: {
|
||||
model: string;
|
||||
quality?: string;
|
||||
@@ -199,6 +235,8 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
outputCostInCents: number;
|
||||
inputTokenCount: number;
|
||||
outputTokenCount: number;
|
||||
outputTextTokenCount: number;
|
||||
outputImageTokenCount: number;
|
||||
}) {
|
||||
const clientDriverCall = Context.get('client_driver_call') as { response_metadata?: Record<string, unknown> } | undefined;
|
||||
const responseMetadata = clientDriverCall?.response_metadata;
|
||||
@@ -218,6 +256,8 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
ratio: `${ratio.w}x${ratio.h}`,
|
||||
input_tokens: inputTokenCount,
|
||||
output_tokens: outputTokenCount,
|
||||
output_text_tokens: outputTextTokenCount,
|
||||
output_image_tokens: outputImageTokenCount,
|
||||
input_microcents: this.#toMicroCents(inputCostInCents),
|
||||
output_microcents: this.#toMicroCents(outputCostInCents),
|
||||
total_microcents: this.#toMicroCents(totalCostInCents),
|
||||
@@ -226,9 +266,27 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
|
||||
#extractUsageMetadata (response: GenerateContentResponse): GeminiUsageMetadata {
|
||||
const usage = (response as GenerateContentResponse & { usageMetadata?: Record<string, unknown> }).usageMetadata;
|
||||
|
||||
let candidatesImageTokenCount = 0;
|
||||
|
||||
const details = usage?.candidatesTokensDetails;
|
||||
if ( Array.isArray(details) ) {
|
||||
for ( const entry of details ) {
|
||||
if ( entry?.modality === 'IMAGE' ) {
|
||||
candidatesImageTokenCount += this.#toSafeCount(entry.tokenCount);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// api only returns modality image, so calculate text tokens as candidates (output) - image tokens
|
||||
const candidatesTokenCount = this.#toSafeCount(usage?.candidatesTokenCount);
|
||||
const candidatesTextTokenCount = Math.max(0, candidatesTokenCount - candidatesImageTokenCount);
|
||||
|
||||
return {
|
||||
promptTokenCount: this.#toSafeCount(usage?.promptTokenCount),
|
||||
candidatesTokenCount: this.#toSafeCount(usage?.candidatesTokenCount),
|
||||
candidatesTokenCount,
|
||||
candidatesTextTokenCount,
|
||||
candidatesImageTokenCount,
|
||||
thoughtsTokenCount: this.#toSafeCount(usage?.thoughtsTokenCount),
|
||||
};
|
||||
}
|
||||
@@ -266,19 +324,44 @@ export class GeminiImageGenerationProvider implements IImageProvider {
|
||||
|
||||
for ( const part of parts ) {
|
||||
if ( part?.inlineData?.data ) {
|
||||
return `data:image/png;base64,${ part.inlineData.data}`;
|
||||
const mimeType = part.inlineData.mimeType ?? 'image/png';
|
||||
return `data:${mimeType};base64,${ part.inlineData.data}`;
|
||||
}
|
||||
}
|
||||
return undefined;
|
||||
}
|
||||
|
||||
#detectMimeType (data: string): string | undefined {
|
||||
// Handle data URIs like "data:image/jpeg;base64,..."
|
||||
const parsed = this.#parseDataUri(data);
|
||||
if ( parsed ) {
|
||||
return parsed.mimeType;
|
||||
}
|
||||
|
||||
for ( const [signature, mimeType] of Object.entries(MIME_SIGNATURES) ) {
|
||||
if ( data.startsWith(signature) ) {
|
||||
return mimeType;
|
||||
}
|
||||
}
|
||||
return undefined;
|
||||
}
|
||||
|
||||
#parseDataUri (data: string): { mimeType: string; base64: string } | undefined {
|
||||
if ( ! data.startsWith('data:image/') ) return undefined;
|
||||
|
||||
const commaIdx = data.indexOf(',');
|
||||
if ( commaIdx === -1 ) return undefined;
|
||||
|
||||
const header = data.substring(5, commaIdx); // after "data:" up to ","
|
||||
if ( ! header.endsWith(';base64') ) return undefined;
|
||||
|
||||
const mimeType = header.substring(0, header.length - 7); // strip ";base64"
|
||||
if ( mimeType.length === 0 ) return undefined;
|
||||
|
||||
return { mimeType, base64: data.substring(commaIdx + 1) };
|
||||
}
|
||||
|
||||
#isValidRatio (ratio: { w: number; h: number }, allowedRatios: { w: number; h: number }[]) {
|
||||
return allowedRatios.some(r => r.w === ratio.w && r.h === ratio.h);
|
||||
}
|
||||
|
||||
#isValidImageMimeType (mimeType?: string) {
|
||||
if ( ! mimeType ) return false;
|
||||
const supportedTypes = ['image/png', 'image/jpeg', 'image/jpg', 'image/webp'];
|
||||
return supportedTypes.includes(mimeType.toLowerCase());
|
||||
}
|
||||
}
|
||||
|
||||
+31
-102
@@ -21,6 +21,22 @@ import { IImageModel } from '../types';
|
||||
|
||||
export const GEMINI_DEFAULT_RATIO = { w: 1024, h: 1024 };
|
||||
|
||||
// Estimated image output token counts for pre-flight cost checks.
|
||||
// These are based on Google's published pricing equivalences.
|
||||
// https://ai.google.dev/gemini-api/docs/image-generation#aspect_ratios_and_image_size
|
||||
export const GEMINI_ESTIMATED_IMAGE_TOKENS: Record<string, number> = {
|
||||
'gemini-2.5-flash-image': 1290,
|
||||
|
||||
'gemini-3-pro-image-preview:1K': 1120,
|
||||
'gemini-3-pro-image-preview:2K': 1120,
|
||||
'gemini-3-pro-image-preview:4K': 2000,
|
||||
|
||||
'gemini-3.1-flash-image-preview:0.5K': 747,
|
||||
'gemini-3.1-flash-image-preview:1K': 1120,
|
||||
'gemini-3.1-flash-image-preview:2K': 1680,
|
||||
'gemini-3.1-flash-image-preview:4K': 2520,
|
||||
};
|
||||
|
||||
export const GEMINI_IMAGE_GENERATION_MODELS: IImageModel[] = [
|
||||
{
|
||||
puterId: 'google:google/gemini-2.5-flash-image',
|
||||
@@ -37,20 +53,13 @@ export const GEMINI_IMAGE_GENERATION_MODELS: IImageModel[] = [
|
||||
version: '1.0',
|
||||
costs_currency: 'usd-cents',
|
||||
index_cost_key: '1x1',
|
||||
index_input_cost_key: 'prompt_tokens',
|
||||
index_input_cost_key: 'input',
|
||||
allowedQualityLevels: [''],
|
||||
costs: {
|
||||
input: 30, // $0.30 per 1M prompt tokens, however google counts them
|
||||
'1x1': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'2x3': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'3x2': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'3x4': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'4x3': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'4x5': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'5x4': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'9x16': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'16x9': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
'21x9': 3.9, // $0.039 per image, just used for extiamte input allowed usage
|
||||
input: 30, // $0.30 per 1M input tokens (text/image)
|
||||
output: 250, // $2.50 per 1M output tokens (text and thinking)
|
||||
output_image: 3000, // $30.00 per 1M output image tokens
|
||||
'1x1': 3.9,
|
||||
},
|
||||
allowedRatios: [
|
||||
{ w: 1, h: 1 },
|
||||
@@ -72,6 +81,7 @@ export const GEMINI_IMAGE_GENERATION_MODELS: IImageModel[] = [
|
||||
version: '1.0',
|
||||
costs_currency: 'usd-cents',
|
||||
index_cost_key: '1K:1x1',
|
||||
index_input_cost_key: 'input',
|
||||
aliases: [
|
||||
'gemini-3-pro-image-preview',
|
||||
'gemini-3-pro-image',
|
||||
@@ -93,38 +103,10 @@ export const GEMINI_IMAGE_GENERATION_MODELS: IImageModel[] = [
|
||||
{ w: 21, h: 9 },
|
||||
],
|
||||
costs: {
|
||||
input: 200, // $2.00 per 1M input tokens, however google counts them
|
||||
output: 1200, // $12.00 per 1M output tokens, however google counts them
|
||||
'1K:1x1': 13.4, // $0.134 per image
|
||||
'1K:2x3': 13.4, // $0.134 per image
|
||||
'1K:3x2': 13.4, // $0.134 per image
|
||||
'1K:3x4': 13.4, // $0.134 per image
|
||||
'1K:4x3': 13.4, // $0.134 per image
|
||||
'1K:4x5': 13.4, // $0.134 per image
|
||||
'1K:5x4': 13.4, // $0.134 per image
|
||||
'1K:9x16': 13.4, // $0.134 per image
|
||||
'1K:16x9': 13.4, // $0.134 per image
|
||||
'1K:21x9': 13.4, // $0.134 per image
|
||||
'2K:1x1': 13.4, // $0.134 per image
|
||||
'2K:2x3': 13.4, // $0.134 per image
|
||||
'2K:3x2': 13.4, // $0.134 per image
|
||||
'2K:3x4': 13.4, // $0.134 per image
|
||||
'2K:4x3': 13.4, // $0.134 per image
|
||||
'2K:4x5': 13.4, // $0.134 per image
|
||||
'2K:5x4': 13.4, // $0.134 per image
|
||||
'2K:9x16': 13.4, // $0.134 per image
|
||||
'2K:16x9': 13.4, // $0.134 per image
|
||||
'2K:21x9': 13.4, // $0.134 per image
|
||||
'4K:1x1': 24, // $0.24 per image
|
||||
'4K:2x3': 24, // $0.24 per image
|
||||
'4K:3x2': 24, // $0.24 per image
|
||||
'4K:3x4': 24, // $0.24 per image
|
||||
'4K:4x3': 24, // $0.24 per image
|
||||
'4K:4x5': 24, // $0.24 per image
|
||||
'4K:5x4': 24, // $0.24 per image
|
||||
'4K:9x16': 24, // $0.24 per image
|
||||
'4K:16x9': 24, // $0.24 per image
|
||||
'4K:21x9': 24, // $0.24 per image
|
||||
input: 200, // $2.00 per 1M input tokens (text/image)
|
||||
output: 1200, // $12.00 per 1M output tokens (text and thinking)
|
||||
output_image: 12000, // $120.00 per 1M output image tokens
|
||||
'1K:1x1': 13.4,
|
||||
},
|
||||
},
|
||||
{
|
||||
@@ -134,6 +116,7 @@ export const GEMINI_IMAGE_GENERATION_MODELS: IImageModel[] = [
|
||||
version: '1.0',
|
||||
costs_currency: 'usd-cents',
|
||||
index_cost_key: '1K:1x1',
|
||||
index_input_cost_key: 'input',
|
||||
aliases: [
|
||||
'gemini-3.1-flash-image-preview',
|
||||
'gemini-3.1-flash-image',
|
||||
@@ -159,64 +142,10 @@ export const GEMINI_IMAGE_GENERATION_MODELS: IImageModel[] = [
|
||||
{ w: 21, h: 9 },
|
||||
],
|
||||
costs: {
|
||||
input: 25, // $0.25 per 1M input tokens, however google counts them
|
||||
output: 150, // $1.50 per 1M output tokens, however google counts them
|
||||
'0.5K:1x1': 4.5, // $0.045 per image
|
||||
'0.5K:1x4': 4.5, // $0.045 per image
|
||||
'0.5K:1x8': 4.5, // $0.045 per image
|
||||
'0.5K:2x3': 4.5, // $0.045 per image
|
||||
'0.5K:3x2': 4.5, // $0.045 per image
|
||||
'0.5K:3x4': 4.5, // $0.045 per image
|
||||
'0.5K:4x1': 4.5, // $0.045 per image
|
||||
'0.5K:4x3': 4.5, // $0.045 per image
|
||||
'0.5K:4x5': 4.5, // $0.045 per image
|
||||
'0.5K:5x4': 4.5, // $0.045 per image
|
||||
'0.5K:8x1': 4.5, // $0.045 per image
|
||||
'0.5K:9x16': 4.5, // $0.045 per image
|
||||
'0.5K:16x9': 4.5, // $0.045 per image
|
||||
'0.5K:21x9': 4.5, // $0.045 per image
|
||||
'1K:1x1': 6.7, // $0.067 per image
|
||||
'1K:1x4': 6.7, // $0.067 per image
|
||||
'1K:1x8': 6.7, // $0.067 per image
|
||||
'1K:2x3': 6.7, // $0.067 per image
|
||||
'1K:3x2': 6.7, // $0.067 per image
|
||||
'1K:3x4': 6.7, // $0.067 per image
|
||||
'1K:4x1': 6.7, // $0.067 per image
|
||||
'1K:4x3': 6.7, // $0.067 per image
|
||||
'1K:4x5': 6.7, // $0.067 per image
|
||||
'1K:5x4': 6.7, // $0.067 per image
|
||||
'1K:8x1': 6.7, // $0.067 per image
|
||||
'1K:9x16': 6.7, // $0.067 per image
|
||||
'1K:16x9': 6.7, // $0.067 per image
|
||||
'1K:21x9': 6.7, // $0.067 per image
|
||||
'2K:1x1': 10.1, // $0.101 per image
|
||||
'2K:1x4': 10.1, // $0.101 per image
|
||||
'2K:1x8': 10.1, // $0.101 per image
|
||||
'2K:2x3': 10.1, // $0.101 per image
|
||||
'2K:3x2': 10.1, // $0.101 per image
|
||||
'2K:3x4': 10.1, // $0.101 per image
|
||||
'2K:4x1': 10.1, // $0.101 per image
|
||||
'2K:4x3': 10.1, // $0.101 per image
|
||||
'2K:4x5': 10.1, // $0.101 per image
|
||||
'2K:5x4': 10.1, // $0.101 per image
|
||||
'2K:8x1': 10.1, // $0.101 per image
|
||||
'2K:9x16': 10.1, // $0.101 per image
|
||||
'2K:16x9': 10.1, // $0.101 per image
|
||||
'2K:21x9': 10.1, // $0.101 per image
|
||||
'4K:1x1': 15.1, // $0.151 per image
|
||||
'4K:1x4': 15.1, // $0.151 per image
|
||||
'4K:1x8': 15.1, // $0.151 per image
|
||||
'4K:2x3': 15.1, // $0.151 per image
|
||||
'4K:3x2': 15.1, // $0.151 per image
|
||||
'4K:3x4': 15.1, // $0.151 per image
|
||||
'4K:4x1': 15.1, // $0.151 per image
|
||||
'4K:4x3': 15.1, // $0.151 per image
|
||||
'4K:4x5': 15.1, // $0.151 per image
|
||||
'4K:5x4': 15.1, // $0.151 per image
|
||||
'4K:8x1': 15.1, // $0.151 per image
|
||||
'4K:9x16': 15.1, // $0.151 per image
|
||||
'4K:16x9': 15.1, // $0.151 per image
|
||||
'4K:21x9': 15.1, // $0.151 per image
|
||||
input: 25, // $0.25 per 1M input tokens (text/image)
|
||||
output: 150, // $1.50 per 1M output tokens (text and thinking)
|
||||
output_image: 6000, // $60.00 per 1M output image tokens
|
||||
'1K:1x1': 6.7,
|
||||
},
|
||||
},
|
||||
];
|
||||
|
||||
@@ -21,6 +21,9 @@ export interface IGenerateParams {
|
||||
provider?: string,
|
||||
test_mode?: boolean
|
||||
quality?: string,
|
||||
input_image?: string,
|
||||
input_image_mime_type?: string,
|
||||
input_images?: string[],
|
||||
};
|
||||
export interface IImageProvider {
|
||||
|
||||
|
||||
Reference in New Issue
Block a user