feat: label image references in generation

This commit is contained in:
stupid-h4er
2026-05-31 00:54:54 +08:00
parent 78738e060f
commit c3b1ba2943
6 changed files with 57 additions and 10 deletions
+1
View File
@@ -1,5 +1,6 @@
# 待测试
- 生图工作台和画布生图会把参考图按当前顺序显示为 `图片1``图片2` 等编号,并在图生图请求的实际提示词中注入编号说明;需要验证 `/image` 参考图排序、画布配置节点输入顺序和画布助手参考图编号一致。
- GPT Image 生图请求会在前端把 `9:16``16:9` 等比例转换成合法 `WIDTHxHEIGHT` 尺寸,并在非法尺寸时直接显示中文错误,避免上游返回 `invalid_value Invalid size`
- Docker 部署时,`DATABASE_DSN=data/infinite-canvas.db` 会在存在 `/app/data` 挂载目录时自动归一到 `/app/data/infinite-canvas.db`,需要验证后台模型配置不会再因为工作目录变为 `/app/web` 而读到空库。
- Seedance 参考视频被火山判定包含真人或隐私信息时,前端错误摘要会提示改用不含真人的视频、官方允许的模型产物或已授权的 `asset://` 素材;参考素材上传目录改为跟随 SQLite 数据目录,并补充公开素材的 HEAD 访问。
@@ -16,6 +16,7 @@ import { requestEdit, requestGeneration, requestImageQuestion, type ChatCompleti
import { imageToDataUrl, uploadImage } from "@/services/image-storage";
import { useAssetStore } from "@/stores/use-asset-store";
import { useThemeStore } from "@/stores/use-theme-store";
import { imageReferenceLabel } from "@/lib/image-reference-prompt";
import type { ReferenceImage } from "@/types/image";
import { DiaTextReveal } from "@/components/ui/dia-text-reveal";
import { CanvasImageSettingsPopover } from "./canvas-image-settings-popover";
@@ -398,8 +399,8 @@ function AssistantComposer({
<div className="px-2 pb-2" onWheelCapture={(event) => event.stopPropagation()}>
{references.length ? (
<div className="thin-scrollbar mb-1.5 flex max-w-full gap-1.5 overflow-x-auto px-1 pb-1">
{references.map((item) => (
<AssistantReferenceChip key={item.id} item={item} onRemove={() => onRemoveReference(item.id)} />
{references.map((item, index) => (
<AssistantReferenceChip key={item.id} item={item} label={assistantImageReferenceLabel(references, index)} onRemove={() => onRemoveReference(item.id)} />
))}
</div>
) : null}
@@ -584,20 +585,23 @@ function AssistantHistory({
function MessageReferences({ message }: { message: CanvasAssistantMessage }) {
return (
<div className={cn("flex max-w-[88%] flex-wrap gap-2", message.role === "user" ? "justify-end" : "justify-start")}>
{message.references?.map((item) => (
<AssistantReferenceChip key={item.id} item={item} />
{message.references?.map((item, index, references) => (
<AssistantReferenceChip key={item.id} item={item} label={assistantImageReferenceLabel(references, index)} />
))}
</div>
);
}
function AssistantReferenceChip({ item, onRemove }: { item: CanvasAssistantReference; onRemove?: () => void }) {
function AssistantReferenceChip({ item, label, onRemove }: { item: CanvasAssistantReference; label?: string; onRemove?: () => void }) {
const theme = canvasThemes[useThemeStore((state) => state.theme)];
const text = (item.text || item.title).replace(/\s+/g, " ").trim().slice(0, 1) || "文";
return (
<div className="group/chip relative inline-flex h-8 max-w-[150px] shrink-0 items-center gap-1.5 rounded-lg text-sm" style={{ color: theme.node.text }}>
{item.dataUrl ? (
<img src={item.dataUrl} alt="" className="size-8 rounded-lg object-cover" />
<span className="relative block size-8 shrink-0">
<img src={item.dataUrl} alt="" className="size-8 rounded-lg object-cover" />
{label ? <span className="absolute left-0.5 top-0.5 rounded bg-black/60 px-1 py-0.5 text-[8px] font-medium leading-none text-white">{label}</span> : null}
</span>
) : (
<span className="grid size-8 place-items-center rounded-lg border text-sm font-medium" style={{ background: theme.node.panel, borderColor: theme.node.activeStroke }}>
{text}
@@ -618,6 +622,12 @@ function AssistantReferenceChip({ item, onRemove }: { item: CanvasAssistantRefer
);
}
function assistantImageReferenceLabel(references: CanvasAssistantReference[], index: number) {
if (!references[index]?.dataUrl) return undefined;
const imageIndex = references.slice(0, index + 1).filter((item) => item.dataUrl).length - 1;
return imageIndex >= 0 ? imageReferenceLabel(imageIndex) : undefined;
}
function nodeToReference(node: CanvasNodeData): CanvasAssistantReference | null {
if (node.type === CanvasNodeType.Image && node.metadata?.content) {
return { id: node.id, type: node.type, title: node.title, dataUrl: node.metadata.content, storageKey: node.metadata.storageKey };
@@ -9,6 +9,7 @@ import { ModelPicker } from "@/components/model-picker";
import { defaultConfig, useConfigStore, useEffectiveConfig, type AiConfig } from "@/stores/use-config-store";
import { CreditSymbol, requestCreditCost } from "@/constant/credits";
import { canvasThemes } from "@/lib/canvas-theme";
import { imageReferenceLabel } from "@/lib/image-reference-prompt";
import { seedanceReferenceLabel } from "@/lib/seedance-video";
import { useThemeStore } from "@/stores/use-theme-store";
import { CanvasImageSettingsPopover } from "./canvas-image-settings-popover";
@@ -304,7 +305,7 @@ function ImageSortCard({
<div className="w-24 shrink-0 overflow-hidden rounded-lg border" style={{ background: theme.node.fill, borderColor: theme.node.stroke }}>
<div className="relative">
<img src={input.image.dataUrl} alt={input.title} className="aspect-square w-full object-cover" />
<span className="absolute left-1 top-1 rounded bg-black/50 px-1 py-0.5 text-[9px] font-medium text-white">{seedanceReferenceLabel("image", imageIndex)}</span>
<span className="absolute left-1 top-1 rounded bg-black/50 px-1 py-0.5 text-[9px] font-medium text-white">{imageReferenceLabel(imageIndex)}</span>
<HorizontalOrderButtons index={imageIndex} total={imageTotal} onMove={(offset) => onMove(input, offset)} />
</div>
</div>
+23 -2
View File
@@ -1,6 +1,6 @@
"use client";
import { BookOpen, CheckSquare, ClipboardPaste, Download, FolderPlus, History, ImagePlus, LoaderCircle, PenLine, Plus, SlidersHorizontal, Sparkles, Trash2, Upload } from "lucide-react";
import { ArrowLeft, ArrowRight, BookOpen, CheckSquare, ClipboardPaste, Download, FolderPlus, History, ImagePlus, LoaderCircle, PenLine, Plus, SlidersHorizontal, Sparkles, Trash2, Upload } from "lucide-react";
import { useEffect, useRef, useState } from "react";
import { App, Button, Checkbox, Drawer, Empty, Image, Input, Modal, Tag, Typography } from "antd";
import localforage from "localforage";
@@ -11,6 +11,7 @@ import { ModelPicker } from "@/components/model-picker";
import { PromptSelectDialog } from "@/components/prompts/prompt-select-dialog";
import { AssetPickerModal, type InsertAssetPayload } from "@/app/(user)/canvas/components/asset-picker-modal";
import { canvasThemes } from "@/lib/canvas-theme";
import { imageReferenceLabel } from "@/lib/image-reference-prompt";
import { useConfigStore, useEffectiveConfig, type AiConfig } from "@/stores/use-config-store";
import { useThemeStore } from "@/stores/use-theme-store";
import { nanoid } from "nanoid";
@@ -373,9 +374,11 @@ export default function ImagePage() {
event.currentTarget.scrollLeft += event.deltaY;
}}
>
{references.map((item) => (
{references.map((item, index) => (
<div key={item.id} className="group relative size-20 shrink-0 overflow-hidden rounded-md border border-stone-200 dark:border-stone-800">
<img src={item.dataUrl} alt={item.name} className="size-full object-cover" />
<span className="absolute left-1 top-1 rounded bg-black/60 px-1.5 py-0.5 text-[10px] font-medium text-white">{imageReferenceLabel(index)}</span>
<ReferenceOrderButtons index={index} total={references.length} onMove={(offset) => setReferences((value) => moveListItem(value, index, offset))} />
<button
type="button"
className="absolute right-1 top-1 hidden size-6 items-center justify-center rounded bg-black/60 text-white group-hover:flex"
@@ -742,6 +745,24 @@ function normalizeLogConfig(log: Partial<GenerationLog>): GenerationLogConfig {
};
}
function moveListItem<T>(items: T[], index: number, offset: number) {
const targetIndex = index + offset;
if (targetIndex < 0 || targetIndex >= items.length) return items;
const next = [...items];
[next[index], next[targetIndex]] = [next[targetIndex], next[index]];
return next;
}
function ReferenceOrderButtons({ index, total, onMove }: { index: number; total: number; onMove: (offset: number) => void }) {
if (total <= 1) return null;
return (
<div className="absolute inset-x-1 bottom-1 flex justify-between">
<Button size="small" className="!h-6 !w-6 !min-w-6 !rounded-full !bg-white/85 !p-0 !shadow-sm" icon={<ArrowLeft className="size-3" />} disabled={index <= 0} onClick={() => onMove(-1)} />
<Button size="small" className="!h-6 !w-6 !min-w-6 !rounded-full !bg-white/85 !p-0 !shadow-sm" icon={<ArrowRight className="size-3" />} disabled={index >= total - 1} onClick={() => onMove(1)} />
</div>
);
}
function buildLog({
prompt,
model,
+12
View File
@@ -0,0 +1,12 @@
import type { ReferenceImage } from "@/types/image";
export function imageReferenceLabel(index: number) {
return `图片${index + 1}`;
}
export function buildImageReferencePromptText(prompt: string, references: ReferenceImage[]) {
const text = prompt.trim();
if (!references.length) return text;
const labels = references.map((_, index) => imageReferenceLabel(index));
return `参考图片编号:${labels.join("、")}。请按这些编号理解提示词中的图片引用。\n\n${text}`;
}
+3 -1
View File
@@ -4,6 +4,7 @@ import { buildApiUrl, type AiConfig } from "@/stores/use-config-store";
import { useUserStore } from "@/stores/use-user-store";
import { nanoid } from "nanoid";
import { dataUrlToFile } from "@/lib/image-utils";
import { buildImageReferencePromptText } from "@/lib/image-reference-prompt";
import { imageToDataUrl } from "@/services/image-storage";
import type { ReferenceImage } from "@/types/image";
@@ -223,9 +224,10 @@ export async function requestEdit(config: AiConfig, prompt: string, references:
const n = Math.max(1, Math.min(15, Math.floor(Math.abs(Number(config.count)) || 1)));
const quality = normalizeQuality(config.quality);
const requestSize = resolveRequestSize(quality, config.size);
const requestPrompt = buildImageReferencePromptText(prompt, references);
const formData = new FormData();
formData.set("model", config.model);
formData.set("prompt", withSystemPrompt(config, prompt));
formData.set("prompt", withSystemPrompt(config, requestPrompt));
formData.set("n", String(n));
formData.set("response_format", "b64_json");
if (quality) {