mirror of
https://github.com/NanmiCoder/cc-haha
synced 2026-07-27 15:13:37 +08:00
feat(plugins): add model capabilities database to image-gen
Agent can now query provider capabilities via list_providers to know: - Supported image sizes per model - Whether the model supports image editing - Whether transparent background is supported - Max number of images per request (n) - Response format (url vs b64_json) Built-in database covers: Agnes image, GPT-image-2, DALL-E 2/3, Gemini image, Flux, Stable Diffusion. Unknown models get pattern-based defaults or "unknown" with full compatibility fallback. generate_image now validates size against capabilities and warns on mismatch. edit_image checks if any provider supports editing before attempting. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
你的姓名
parent
4c128e0764
commit
da1137e468
@ -19,6 +19,40 @@ function isUnset(val) {
|
||||
return !val || val.trim() === '' || val.startsWith('${user_config.')
|
||||
}
|
||||
|
||||
const MODEL_CAPABILITIES = {
|
||||
// Agnes image models
|
||||
'agnes-image-2.1-flash': { sizes: ['512x512', '768x768', '1024x1024'], edit: false, transparent: false, maxN: 4, format: 'url' },
|
||||
'agnes-image-2.0-flash': { sizes: ['512x512', '768x768', '1024x1024'], edit: false, transparent: false, maxN: 4, format: 'url' },
|
||||
// GPT image models
|
||||
'gpt-image-2': { sizes: ['1024x1024', '1536x1024', '1024x1536', 'auto'], edit: true, transparent: true, maxN: 10, format: 'b64_json', notes: 'size must be multiple of 16, max 3840px' },
|
||||
'gpt-image-1': { sizes: ['1024x1024', '1536x1024', '1024x1536', '256x256', '512x512', 'auto'], edit: true, transparent: true, maxN: 10, format: 'b64_json' },
|
||||
// DALL-E models
|
||||
'dall-e-3': { sizes: ['1024x1024', '1792x1024', '1024x1792'], edit: false, transparent: false, maxN: 1, format: 'url' },
|
||||
'dall-e-2': { sizes: ['256x256', '512x512', '1024x1024'], edit: true, transparent: false, maxN: 10, format: 'url' },
|
||||
// Gemini image
|
||||
'gemini-2.5-flash-image-preview': { sizes: ['1024x1024', '512x512', '1536x1536'], edit: true, transparent: false, maxN: 4, format: 'b64_json' },
|
||||
'gemini-2.0-flash-exp-image-generation': { sizes: ['1024x1024', '512x512'], edit: false, transparent: false, maxN: 4, format: 'b64_json' },
|
||||
// Flux models
|
||||
'flux-schnell': { sizes: ['512x512', '768x768', '1024x1024', '1536x1024', '1024x1536'], edit: false, transparent: false, maxN: 4, format: 'url' },
|
||||
'flux-pro': { sizes: ['512x512', '768x768', '1024x1024', '1536x1024', '1024x1536'], edit: false, transparent: true, maxN: 4, format: 'url' },
|
||||
// Stable Diffusion
|
||||
'stable-diffusion-xl': { sizes: ['512x512', '768x768', '1024x1024'], edit: true, transparent: false, maxN: 4, format: 'url' },
|
||||
}
|
||||
|
||||
function getModelCapabilities(model) {
|
||||
// Exact match
|
||||
if (MODEL_CAPABILITIES[model]) return MODEL_CAPABILITIES[model]
|
||||
// Partial match
|
||||
const lower = model.toLowerCase()
|
||||
for (const [key, caps] of Object.entries(MODEL_CAPABILITIES)) {
|
||||
if (lower.includes(key) || key.includes(lower)) return caps
|
||||
}
|
||||
// Pattern-based defaults
|
||||
if (/image/i.test(lower)) return { sizes: ['512x512', '1024x1024'], edit: false, transparent: false, maxN: 4, format: 'url' }
|
||||
// Unknown
|
||||
return null
|
||||
}
|
||||
|
||||
function loadProviderFromEnv(prefix) {
|
||||
const name = process.env[`${prefix}_NAME`]
|
||||
const baseUrl = process.env[`${prefix}_BASE_URL`]
|
||||
@ -34,6 +68,7 @@ function loadProviderFromEnv(prefix) {
|
||||
model,
|
||||
enabled: true,
|
||||
timeoutMs: 300_000,
|
||||
capabilities: getModelCapabilities(model),
|
||||
}
|
||||
}
|
||||
|
||||
@ -435,10 +470,19 @@ async function handleToolCall(name, args) {
|
||||
if (!args.prompt || typeof args.prompt !== 'string' || args.prompt.trim() === '') {
|
||||
return { content: [{ type: 'text', text: 'Error: prompt is required and must be a non-empty string.' }], isError: true }
|
||||
}
|
||||
const n = Math.min(Math.max(Math.floor(Number(args.n) || 1), 1), 10)
|
||||
const firstCaps = providers[0]?.capabilities
|
||||
const maxN = firstCaps?.maxN || 10
|
||||
const n = Math.min(Math.max(Math.floor(Number(args.n) || 1), 1), maxN)
|
||||
const requestedSize = args.size || '1024x1024'
|
||||
if (firstCaps && firstCaps.sizes.length > 0 && !firstCaps.sizes.includes(requestedSize) && requestedSize !== 'auto') {
|
||||
return {
|
||||
content: [{ type: 'text', text: `Size "${requestedSize}" may not be supported by ${providers[0].model}. Supported sizes: ${firstCaps.sizes.join(', ')}. Will attempt anyway with compatibility fallback.` }],
|
||||
isError: false,
|
||||
}
|
||||
}
|
||||
const result = await generateWithFallback(
|
||||
args.prompt.trim(),
|
||||
args.size || '1024x1024',
|
||||
requestedSize,
|
||||
n,
|
||||
args.transparent,
|
||||
providers,
|
||||
@ -447,13 +491,26 @@ async function handleToolCall(name, args) {
|
||||
}
|
||||
|
||||
case 'edit_image': {
|
||||
if (!args.prompt || typeof args.prompt !== 'string' || args.prompt.trim() === '') {
|
||||
return { content: [{ type: 'text', text: 'Error: prompt is required.' }], isError: true }
|
||||
}
|
||||
if (!args.image_url || typeof args.image_url !== 'string') {
|
||||
return { content: [{ type: 'text', text: 'Error: image_url is required.' }], isError: true }
|
||||
}
|
||||
const editProviders = providers.filter(p => p.capabilities?.edit !== false)
|
||||
if (editProviders.length === 0) {
|
||||
return {
|
||||
content: [{ type: 'text', text: 'None of the configured providers support image editing. Use generate_image instead, or add a provider that supports editing (e.g., gpt-image-2, gemini-2.5-flash-image-preview).' }],
|
||||
isError: true,
|
||||
}
|
||||
}
|
||||
const result = await editWithFallback(
|
||||
args.prompt,
|
||||
args.prompt.trim(),
|
||||
args.image_url,
|
||||
args.size,
|
||||
args.n,
|
||||
args.size || '1024x1024',
|
||||
args.n || 1,
|
||||
args.transparent,
|
||||
providers,
|
||||
editProviders,
|
||||
)
|
||||
return formatResult(result)
|
||||
}
|
||||
@ -461,7 +518,21 @@ async function handleToolCall(name, args) {
|
||||
case 'list_providers': {
|
||||
const lines = providers.map((p, i) => {
|
||||
const status = p.enabled === false ? ' [DISABLED]' : ''
|
||||
return `${i}. ${p.name}${status}\n baseUrl: ${p.baseUrl}\n model: ${p.model}\n timeout: ${p.timeoutMs || 300_000}ms`
|
||||
const caps = p.capabilities
|
||||
let capStr = ''
|
||||
if (caps) {
|
||||
const parts = []
|
||||
parts.push(`sizes: ${caps.sizes.join(', ')}`)
|
||||
if (caps.edit) parts.push('supports edit')
|
||||
if (caps.transparent) parts.push('supports transparent')
|
||||
parts.push(`max n: ${caps.maxN}`)
|
||||
parts.push(`returns: ${caps.format}`)
|
||||
if (caps.notes) parts.push(`note: ${caps.notes}`)
|
||||
capStr = `\n capabilities: ${parts.join(' | ')}`
|
||||
} else {
|
||||
capStr = '\n capabilities: unknown (will try all params, fallback on error)'
|
||||
}
|
||||
return `${i}. ${p.name}${status}\n baseUrl: ${p.baseUrl}\n model: ${p.model}${capStr}`
|
||||
})
|
||||
return {
|
||||
content: [{ type: 'text', text: `Configured providers (${providers.length}):\n\n${lines.join('\n\n')}` }],
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user