John @johnvirality - Twitter Profile

i've been running Sonnet 5.5 + Jev scoring + Higgsfield + Capcut MCP as ONE closed loop instead of a pile of tools... and it has been ABSOLUTELY cooking most people open a chat window, ask for 10 tiktok ideas, make one, do 400 views, then decide AI content doesn't work generation was never the bottleneck. it's the cheapest part of the process and everybody optimised it first so i documented the ENTIRE machine... covering research, ICP excavation, idea scoring, image gen, video animation, the edit layer, distribution, and the feedback loop that makes next week's batch smarter than this week's here's what's inside the guide: → the context engine that makes every downstream prompt actually work (sales calls, DMs, support tickets, closed-lost notes, 90 days of analytics. the system tells you what context it's MISSING, which is usually how you find out your positioning is vague) → ICP excavation that builds your hook library from your customers' mouths (verbatim quotes mapped to pain, failed solutions, objection, awareness stage. any phrase appearing in 3+ separate conversations gets flagged as a probable hook) → the research layer: platform scrape + adjacent market mining + competitor gap teardown (your competitors are all watching each other. the formats that feel fresh in your category were invented in a different one. this prompt steals the mechanics, not the topics) → the scoring layer that decides what deserves production (50 ideas scored across viral, business and execution dimensions. you produce the best 5 of 50 instead of the first 5 you thought of, for the exact same production cost. the part everyone skips) → the sonnet 5.5 x gpt images 2.0 x seedance 2.5 x higgsfield production stack (consistency lock built once and pasted into every prompt. character reference locked across every shot AND every video in the batch. broken renders flagged and regenerated automatically) → the capcut MCP edit layer, because editing is retention not polish (cut on action never after it, no shot over 2.5s in the first 6 seconds, word-by-word captions, hook in frame 1, CTA in the final 2 seconds) → the closed-loop feedback system that compounds (post-mortem names the exact second retention dropped, winners get doubled down on with one variable held constant, context file updates every week. week 12 you're generating from your own performance history instead of guessing) 19 prompts. written out in full. copy-paste and run. all backed by what i've seen running distribution for 200k+ creators and watching which clips actually convert to installs like + comment "LOOP" and i'll send the whole protocol over (must be following + RT for priority access) will stop sending in 24hrs...

johnvirality's tweet photo. i've been running Sonnet 5.5 + Jev scoring + Higgsfield + Capcut MCP as ONE closed loop instead of a pile of tools...

and it has been ABSOLUTELY cooking

most people open a chat window, ask for 10 tiktok ideas, make one, do 400 views, then decide AI content doesn't work

generation was never the bottleneck. it's the cheapest part of the process and everybody optimised it first

so i documented the ENTIRE machine...

covering research, ICP excavation, idea scoring, image gen, video animation, the edit layer, distribution, and the feedback loop that makes next week's batch smarter than this week's

here's what's inside the guide:

→ the context engine that makes every downstream prompt actually work

(sales calls, DMs, support tickets, closed-lost notes, 90 days of analytics. the system tells you what context it's MISSING, which is usually how you find out your positioning is vague)

→ ICP excavation that builds your hook library from your customers' mouths

(verbatim quotes mapped to pain, failed solutions, objection, awareness stage. any phrase appearing in 3+ separate conversations gets flagged as a probable hook)

→ the research layer: platform scrape + adjacent market mining + competitor gap teardown

(your competitors are all watching each other. the formats that feel fresh in your category were invented in a different one. this prompt steals the mechanics, not the topics)

→ the scoring layer that decides what deserves production

(50 ideas scored across viral, business and execution dimensions. you produce the best 5 of 50 instead of the first 5 you thought of, for the exact same production cost. the part everyone skips)

→ the sonnet 5.5 x gpt images 2.0 x seedance 2.5 x higgsfield production stack

(consistency lock built once and pasted into every prompt. character reference locked across every shot AND every video in the batch. broken renders flagged and regenerated automatically)

→ the capcut MCP edit layer, because editing is retention not polish

(cut on action never after it, no shot over 2.5s in the first 6 seconds, word-by-word captions, hook in frame 1, CTA in the final 2 seconds)

→ the closed-loop feedback system that compounds

(post-mortem names the exact second retention dropped, winners get doubled down on with one variable held constant, context file updates every week. week 12 you're generating from your own performance history instead of guessing)

19 prompts. written out in full. copy-paste and run.

all backed by what i've seen running distribution for 200k+ creators and watching which clips actually convert to installs

like + comment "LOOP" and i'll send the whole protocol over

(must be following + RT for priority access)

will stop sending in 24hrs...

36

33

12

39

2K

John

@johnvirality

3 days ago

@Oussama_buildin Just sent it

0

1

0

17

John

@johnvirality

3 days ago

@fintwit74 Sent

0

9

John

@johnvirality

3 days ago

@therflawless Sent

0

8

John

@johnvirality

3 days ago

@SyedFahimAnwer1 Sent it

0

11

John

@johnvirality

3 days ago

@Sadanpan3 Sent it

0

16

John

@johnvirality

4 days ago

@M_OmerFarooq Sent it

0

11

John

@johnvirality

4 days ago

@claudiofricks Sent

0

8

John

@johnvirality

4 days ago

@ishitaa_rao Just sent it

0

10

John

@johnvirality

4 days ago

how life feels after getting this doc for free...

John

@johnvirality

5 days ago

i've been running Sonnet 5.5 + Jev scoring + Higgsfield + Capcut MCP as ONE closed loop instead of a pile of tools... and it has been ABSOLUTELY cooking most people open a chat window, ask for 10 tiktok ideas, make one, do 400 views, then decide AI content doesn't work generation was never the bottleneck. it's the cheapest part of the process and everybody optimised it first so i documented the ENTIRE machine... covering research, ICP excavation, idea scoring, image gen, video animation, the edit layer, distribution, and the feedback loop that makes next week's batch smarter than this week's here's what's inside the guide: → the context engine that makes every downstream prompt actually work (sales calls, DMs, support tickets, closed-lost notes, 90 days of analytics. the system tells you what context it's MISSING, which is usually how you find out your positioning is vague) → ICP excavation that builds your hook library from your customers' mouths (verbatim quotes mapped to pain, failed solutions, objection, awareness stage. any phrase appearing in 3+ separate conversations gets flagged as a probable hook) → the research layer: platform scrape + adjacent market mining + competitor gap teardown (your competitors are all watching each other. the formats that feel fresh in your category were invented in a different one. this prompt steals the mechanics, not the topics) → the scoring layer that decides what deserves production (50 ideas scored across viral, business and execution dimensions. you produce the best 5 of 50 instead of the first 5 you thought of, for the exact same production cost. the part everyone skips) → the sonnet 5.5 x gpt images 2.0 x seedance 2.5 x higgsfield production stack (consistency lock built once and pasted into every prompt. character reference locked across every shot AND every video in the batch. broken renders flagged and regenerated automatically) → the capcut MCP edit layer, because editing is retention not polish (cut on action never after it, no shot over 2.5s in the first 6 seconds, word-by-word captions, hook in frame 1, CTA in the final 2 seconds) → the closed-loop feedback system that compounds (post-mortem names the exact second retention dropped, winners get doubled down on with one variable held constant, context file updates every week. week 12 you're generating from your own performance history instead of guessing) 19 prompts. written out in full. copy-paste and run. all backed by what i've seen running distribution for 200k+ creators and watching which clips actually convert to installs like + comment "LOOP" and i'll send the whole protocol over (must be following + RT for priority access) will stop sending in 24hrs...

36

33

12

39

2K

11

1

0

8

494