GLM-5.3flash挺有实力的

TomMao23 2026-08-31 10:55 1

Zcode里用下来感觉快速、聪明、而且多模态很好用


附海盗船挑战

GLM5.3flash:两次提示,第一次有穿模bug,第二次提示后完美修复


Gemini3.7flash:一次提示


感兴趣的可以用自己的模型试试

提示词,来自X:

Create from scratch a fully interactive 3D scene of a pirate ship sailing across a dynamic ocean at sunset. The visual style should be cinematic and stylized rather than photorealistic, while still being exceptionally rich, detailed, polished, and visually sophisticated.


Build the ship, ocean, sky, lighting, materials, sails, rigging, cannons, small structural details, sea foam, wake, particles, animation, camera work, composition, atmospheric depth, and color grading. The final result should feel like a premium, high-production 3D artwork, not a prototype, technical demo, or low-quality scene.


Start from a completely blank page. Do not reuse or depend on any previous project or scene. You may create the assets yourself or use reliable, trusted, open-source assets and libraries when necessary.


Mandatory requirements:


No text of any kind may appear anywhere inside the scene. No titles, names, logos, descriptions, credits, labels, or control instructions in any language.


Deliver the entire project as one final standalone page file that can be opened directly in a web browser, with assets embedded inside it as much as reasonably possible.


The ocean, ship, sails, and camera must all be animated naturally and smoothly. Avoid artificial slow motion or sluggish movement. The ship should feel like it is genuinely moving through the water.


Do not rely on primitive geometric shapes as the finished result. Build a visually convincing and detailed pirate ship, including a carefully shaped hull, masts, sails, rigging, ropes, cannons, railings, lanterns, deck structures, and clearly visible small-scale details.


Lighting must reveal the ship’s geometry and materials clearly. Create a rich sunset atmosphere, deep ocean shading, reflections, convincing sea foam, and a detailed sailing wake behind and around the vessel.


Maintain a strong balance between visual quality and real-time performance, preserving smooth interaction and animation without making an obvious sacrifice in quality.


Automatically use the most appropriate skills, tools, libraries, techniques, and available assets needed to achieve the best result. Do not wait for me to specify which technologies to use.


Actually test the finished result in a desktop web browser. Capture visual screenshots, inspect the browser console for errors, and fix every visible or technical issue you find, including distorted geometry, black screens, failed asset loading, broken animation, poor composition, rendering artifacts, or camera problems.


At the end, verify that the final file opens and works directly, that the scene contains no text whatsoever, and that there are no remaining runtime or loading errors. Then finish the task with only a brief response.

最新回复 (4)
  • 小歪 08-31 11:29
    1



    感觉还是哈基米更好看啊

  • Yiyabo! 08-31 13:43
    2

    周末不是送了3e tokens嘛?我放在我的生产项目上用,差不多花了1.6e tokens。30W行的项目可以很好的进行开发,缓存命中超级高,修复也到位。而且测试做的很充分,也能够理解我的“从用户的角度去进行测试”这种稍微抽象一点的说法。无论是任务定时设置,还是意图理解都很到位。还有就是超级快哇,又快又准!


    这次我真能给到夯了。fable5 + 5.3flash就是我目前用着最舒服的组合。我这边Claude和GPT模型是畅用,并且也自主部署了GLM。团队里面的同学基本都认可fable+glm这套操作

  • Yiyabo! 08-31 13:45
    3

    再补充一句,GLM指令遵循非常好,比opus强多了。用GLM+fable的好处就是,fable给出来的计划GLM会非常忠实的去执行。做完后fable再审核。opus问题其实很大,就是做着做着会跑偏,如果没有人盯着,不知道啥时候就偏离了。在科研上这个属于大忌讳

  • 依然犯特困 08-31 13:59
    4

    现在这个模型的价格如何,性价高吗

* 帖子来源Linux.do
返回