大家有没有发现100美元的codex确实不耐用，测试了一下xiaomi mimo 2.5 pro,感觉还不错，大家用了吗？
中文
Alex | WaveShift
106 posts
Alex | WaveShift
@jbang20042004
Building WaveShift — AI video translation, dubbing, and synced subtitles. Make every video multilingual.
Katılım Ocak 2025
142 Takip Edilen7 Takipçiler
I don’t care how many followers you have.
If you’re active, you’re into AI, and you’re a decent human being, I support you!
English
@SRaghuwans87264 Building WaveShift — AI video translation, dubbing, and synced subtitles. Make every video multilingual.
waveshift.net
English
Founders, what are you working on?
Drop your SaaS below:
- What it does
- Who it’s for
Let’s see what you’re building 🚀
English
@sophie_launch Building WaveShift — AI video translation, dubbing, and synced subtitles. Make every video multilingual.
waveshift.net
English
Prompt I use when downtime is allowed:
“Before making changes, do not assume this system requires zero-downtime deployment.
For this task, a planned maintenance window is allowed.
You may propose breaking internal changes if they lead to a cleaner long-term architecture.
Do not preserve legacy compatibility unless it is truly necessary.
Please compare two plans:
Zero-downtime compatible plan
additive changes
adapters
fallback paths
phased migration
legacy compatibility
Clean-break maintenance-window plan
remove obsolete code
simplify data models
rename bad abstractions
consolidate duplicated paths
update schemas directly if appropriate
provide backup, migration, and verification steps
Then recommend the better route.
Constraints:
preserve user data unless explicitly allowed otherwise
preserve external APIs unless explicitly allowed to break them
list all breaking changes before implementation
include rollback or recovery steps
include tests or verification commands”
English
A small prompt change that improves AI refactors:
Tell the agent whether downtime is allowed.
If you don’t, it often assumes zero-downtime compatibility.
So it adds adapters, fallbacks, duplicate paths, and legacy-preserving code.
Safe, but dirty.
For early products, prototypes, internal tools, or systems with a maintenance window, say:
“Downtime is allowed.
Do not optimize for zero-downtime compatibility.
Prefer the cleanest long-term architecture.
List all breaking changes before implementation.”
This makes the agent stop asking:
“How do I preserve every old path?”
And start asking:
“What should be deleted, renamed, collapsed, or redesigned?”
Sometimes the most important prompt is not a feature request.
It is a constraint reset.
Prompt below.
English
My favorite ending from the tutorial:
“What you see isn't just AI. It's the movies I've watched, the music I've heard, the books I've read, and the words I've written.”
That may be the real reason the short worked.
Alex | WaveShift@jbang20042004
The viral AI short “Zombie Scavenger / Zombie Cleaner” finally has an English making-of breakdown. 42 min: prompts → shots, consistency, motion control, editing rhythm. I translated and dubbed it with WaveShift. Original creator: Mx-Shell.
English
@jbang20042004 @rdbuilds7 U can make like a something different from others like a person can replace the person in video to another person with voiceover so it will be good
English
Tiny post-production choices made this AI action scene feel more physical:
- subtle speed ramps
- matching motion direction
- keeping light direction consistent
- mirroring shots when needed
Most people focus on prompts. He focuses on continuity.
Alex | WaveShift@jbang20042004
The viral AI short “Zombie Scavenger / Zombie Cleaner” finally has an English making-of breakdown. 42 min: prompts → shots, consistency, motion control, editing rhythm. I translated and dubbed it with WaveShift. Original creator: Mx-Shell.
English
@sudheerdotai @FlippedRay Thanks for sharing，it looks nice😁
English
English
Prompt I use before major AI refactors:
“Do not refactor yet.
First, audit this project as a senior architect.
If we wanted to refactor it according to the ideal architecture, what substrate is missing today?
Please identify:
missing tests
unclear module boundaries
hidden business rules
unstable contracts
weak or missing types
missing docs
missing invariants
missing validation scripts
missing repo-level instructions
Then give me:
A. why a direct refactor would only produce a local optimum
B. what substrate must be fixed first
C. what should not be touched yet
D. a staged plan for preparing the codebase
E. the safest first PR”
English
A useful trick for AI refactors:
Don’t ask the agent to refactor first.
Ask:
“What makes the ideal refactor impossible in this codebase today?”
That question changes everything.
The agent stops optimizing inside the current mess and starts identifying the missing substrate:
tests
boundaries
contracts
types
docs
invariants
validation loops
repo instructions
Fix the substrate first.
Then refactor.
Context debt comes before code debt.
Prompt below.
English
