dreamina-visual-target
Skill0 starsChanged 23 days agoApache-2.0View on GitHub
Obtain the target image for a visual-quality loop when the user has not supplied one — generate it with the approved paid image tool, or baseline-refine from an existing artifact so the target converges instead of diverging. Use when the user asks to "iterate to a target", has no reference image, says "帮我定个目标图", "生成目标图", "对标图", "target image", "goal image", or when a visual loop needs a target but `lock_target` has nothing to lock. Enforces the anti-concept-art discipline (the target is a screenshot-exact rendering to compare against, NOT a stylized interpretation) and fail-closed behavior without paid-generation approval. NEVER generates without the user's explicit approval — target generation is a paid submission. Do not use for judging (that is `dreamina-vision-judge`).
In this repository30 other files it gives its agents.
- AGENTS.mdAGENTS.md
- openspec-apply-changeSkill
- openspec-archive-changeSkill
- openspec-exploreSkill
- openspec-proposeSkill
- openspec-sync-specsSkill
- openspec-update-changeSkill
- dreamina-auto-seedanceSkill
- dreamina-cli-image2imageSkill
- dreamina-cli-image2videoSkill
- dreamina-cliSkill
- dreamina-cli-text2imageSkill
- dreamina-cli-text2videoSkill
- dreamina-design-harnessSkill
- dreamina-design-useSkill
- dreamina-opencli-image2imageSkill
- dreamina-opencli-image2videoSkill
- dreamina-opencli-text2imageSkill
- dreamina-opencli-text2videoSkill
- dreamina-prompt-image2imageSkill
- dreamina-prompt-image2videoSkill
- dreamina-prompt-text2imageSkill
- dreamina-prompt-text2videoSkill
- dreamina-seedance-resumeSkill
- dreamina-shot-annotatorSkill
- dreamina-video-evaluatorSkill
- dreamina-video-productionSkill
- dreamina-video-shotsSkill
- dreamina-video-syncSkill
- dreamina-vision-judgeSkill
The file
Files
dreamina-visual-target — 目标素材获取
为视觉质量循环取得目标素材。目标是循环的对标锚点:评审按它与候选图比对,
目标的质量直接决定迭代上限。本技能只负责"把一个合格的目标锁定进循环",
不负责评审(评审是 dreamina-vision-judge),也不负责执行生成轮次。
三条获取路径
按顺序判断,一次只走一条:
路径 A:用户已提供目标
用户给了目标图 → 直接校验后进入 lock_target,不要再生成。
这是唯一零付费的路径。
路径 B:已有产物,缺目标(基线精修)
用户已有一个产物或早期版本,但没给目标:
- 取得当前产物的截图或文件作为输入;
- 用图像生成(
dreamina_submit_image,付费)以当前产物为输入生成改进版目标; - 目标必须保留原产物的主体与构图方向——是收敛,不是另一个方向。
跳过基线精修直接生成全新目标,会让目标与现有产物在主体、构图上漂移, 评审的四个维度同时大幅偏离,迭代难度反而更高。
路径 C:全新创作,无目标
不存在任何现有产物 → 直接用图像生成(付费)产出全新目标, 不需要先造一个基线。
反演绎纪律(必须遵守)
目标是可逐项对照的成品呈现,不是风格化诠释:
- ❌ 反例:"一张概念艺术风格的奇幻城堡,氛围宏大" —— 这是艺术演绎, 评审无法用它逐像素或逐区域比对;
- ✅ 正例:"该城堡场景在最终引擎内的实拍截图:正面机位、主体居中、 石材质感清晰、黄昏侧光" —— 构图、光照、材质、细节四个维度都可对照。
生成意图描述中避免把结果推向演绎风格的表述。纪律约束的是意图类别 (可对照成品 vs 风格化诠释),不维护禁用词表。
目标质量自检(锁定前)
- 目标与候选处于同一呈现类别(都是成品呈现,不是一个概念图对一个成品)
- 画幅可比(aspect 一致或接近,评审的构图维度才能比对)
- 逐项可对照:能指着目标的某个区域说出"这里应该是 X"
付费门禁(不可绕)
目标生成是一次付费提交,与 dreamina-design-harness §2 同一门禁:
- 必须先获得用户对该次生成的明确授权;
- 无授权、审批被拒、图像生成能力不可用 → 停下询问, 不静默降级为"无目标迭代",不借用其他通道;
- 不自动重试。
来源记录(锁定时)
lock_target 的 source 字段必须如实记录:
{"kind": "user_supplied"}
{"kind": "generated", "provider": "dreamina", "submit_id": "<真实提交标识>"}
{"kind": "existing_artifact"}生成类目标必须携带真实 submit_id,使目标可审计。目标一经锁定不可被静默替换;
换目标 = 显式创建新目标版本。
与其他技能的衔接
- 拿到目标后 →
dreamina_visual_loop的lock_target锁定,再走run_first_round→dreamina-vision-judge评审 →record_judgement; - 评审结论如何交给修复实现 → 见 harness 的
references/repair-delegation.md。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

