Igor Trunin (@igortr_) header
Igor Trunin (@igortr_) profile picture

Grew Whisper AI to 470K users. Zero paid ads. I test AI tools and tell you what's worth your time. CMO @HypeScribe

Joined September 2023
Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)Photo by Igor Trunin (@igortr_)
ok I have to say it: Opus 5.5 made this whole video 🤯 no After Effects, no editor. I described what I wanted, it wrote every scene in code, rendered it and synced it to the music on its own still can't process that I can just make ads for our product like this now
Nobody reads meeting notes. So we made everything go where your team already works: → recap to Slack, Telegram, Gmail. → action items to Linear, Jira, GitHub, Asana, Trello, Todoist, Motion, Notion. → every call logged in Google Sheets and Airtable, HubSpot, Notion, Google Drive. One click, a chat message, or autopilot. @hypescribe connects to 18 apps, plus MCP for Claude and ChatGPT. Turn on auto-send. Connect Google Calendar and the notetaker joins your Zoom, Meet and Teams calls by itself. The notes go out as soon as the transcript is ready. MCP: search recordings, read transcripts and start new transcriptions from Claude, ChatGPT, Claude Code, Codex or Cursor. API keys for your own scripts. Works on any recording, not just calls: uploads, YouTube links, lectures. Try it below.
3
6
280
Nobody reads meeting notes. So we made everything go where your team already works: → recap to Slack, Telegram, Gmail. → action items to Linear, Jira, GitHub, Asana, Trello, Todoist, Motion, Notion. → every call logged in Google Sheets and Airtable, HubSpot, Notion, Google Drive. One click, a chat message, or autopilot. @hypescribe connects to 18 apps, plus MCP for Claude and ChatGPT. Turn on auto-send. Connect Google Calendar and the notetaker joins your Zoom, Meet and Teams calls by itself. The notes go out as soon as the transcript is ready. MCP: search recordings, read transcripts and start new transcriptions from Claude, ChatGPT, Claude Code, Codex or Cursor. API keys for your own scripts. Works on any recording, not just calls: uploads, YouTube links, lectures. Try it below.
24
4
12
42
16,974
Most of the web still has no API: government portals, insurance directories, anything behind a login. AgentWeb ran 47 real tasks on sites like that, 25 of them signed in, and finished all 47 with a 1.9s median and $3 total. Curious how it holds up on messier workflows.
Introducing AgentWeb: reliable web infrastructure for generalized agents. 47/47 real world tasks on live sites. 1.9s median. $3 total. Request access: agentweb.us
1
5
191
Using ChatGPT as a full-fledged 8K upscaler😌 Don't confuse this with standard image generation; if you simply upload an image, it might get sharper, but the resolution stays the same. The trick here is different. The agent breaks the image into nine parts, upscales each one individually using GPT Image 2, and then stitches them back together into a single whole. Sure, standard upscalers do the exact same thing, but ChatGPT has the advantage of superior visual perception and vast knowledge of the real world. Prompt: --- name: flowers-upscale description: Photo to 8K; 2/3-pass contextual AI crop enhancement and registered blending. --- INPUT S=uploaded photo; missing→request. Unspecified→ask "2 passes (recommended) or 3? 8K default; another resolution?" Await choice. N=2|3; L=specified longest edge else8192; preserve aspect. Discover available AI image-editing+raster-processing capabilities; missing→report requirement. Use environment-native tools/paths; create helpers as needed. Preserve S; separate outputs. Execute end-to-end after choice. E(C): inspect; AI-edit inputs=[S=context,C=target], maximum practical supported resolution; retain actual output dimensions. Prompt=Q+only relevant material constraints. Concurrent independent calls where supported; persist outputs+source-coordinate boxes+prompts immediately. P1: S→3×3 cores; pad each side20% core dimension, clip to S→9 overlapping C→E→register/blend→M1; resize assembled M1 to L. P2: M1→4–8 padded salient ROIs(face/hair/hands/objects/textiles as present)→E→register/blend→M2. Exclude low-information regions. N=3: M2→4–8 tighter focal ROIs(individual eyes/lips/nose/fingers/material details as present)→E→register/blend→M3. Keep homologous features consistent. Every E receives S, never crop alone. Q="1=original: identity/anatomy/color/light authority. 2=target: return exact crop/framing/geometry. Reconstruct plausible photographic detail; preserve expression, pose, contours, clothing, genuine imperfections, focus falloff. Optical clarity; natural edge transitions; spatially varied texture. Repair inherited artifacts. Skin: irregular pores/fine creases, quiet intervals. Hair: cohesive groups, selective fibers, natural sparse fringe. Iris: source color, irregular branching fibers, coherent pupil, restrained reflections. Textiles: original weave direction/spacing/motifs. Avoid invented structures/marks, beautification, relighting, halos, ringing, embossed/crosshatched/repeated texture, synthetic grain, blanket sharpening." ASSEMBLE: map recorded boxes→master; global similarity registration first; bounded smooth local correction only with reliable correspondence, never distort rigid geometry/repeating patterns. Unreliable→retain prior pixels. Match broad exposure/color without transferring old fine texture. Normalized overlap/multiband blend; feather15–25% border→alpha0 at interior patch boundaries. Preserve native enhanced crops. Before next pass: inspect full image+100% focal/overlap crops; repair doubling/seams/drift/texture mismatch. ≤2 targeted retries/defect; rejected/failed→retain prior content+disclose. EXPORT: M_N→lossless PNG at L; verify dimensions+decode; deliver file+preview+prompts/settings. Describe detail as reconstructed; distinguish exported resolution from native generated resolution. Also, an important point: run it in Codex/Work rather than the standard chat, and bump the effort level up to at least "high." On the downside, it takes *forever* to think. The first image took me 10 minutes, and each subsequent one took even longer than the last 😐 You can also use any other image-generating agent instead of GPT - like Gemini + Nanobanana - though the result depends heavily on the agent's "brains." I got the best quality using the Opus 5.5 + Codex CLI image gen combo (a photo of a young Elon), but that was also the longest run by far 😁
2
7
1,453
So I converted his idea into a promt, you can generate any topic, not just people. Prompt: [naive hand-drawn human doodle faces] Write one self-contained HTML file with no libraries and no image assets that uses canvas 2D to draw a grid of 40 variations of the subject in brackets above on textured off-white paper: give each one its own seeded PRNG (same seed = same drawing, click one to reroll it, one button regenerates the whole sheet); start by breaking the subject into 6-12 named parts, each a tiny function with several variants assembled from arcs and curves (that granularity means an eye is two arcs plus a pupil, a nose is one long curve around a tip); then put a rough 3D armature under the 2D art — the simplest solid the subject implies, an ellipsoid for a head or a body, a tilted disc for anything radial, a curve for anything that grows — give it random yaw/pitch/roll and pin every part to a (u,v) anchor on it, projecting that anchor to 2D for position but taking each part's orientation and scale from the armature's own global axes rather than the surface tangent (or the parts converge at the poles), and hide parts whose surface normal faces away; draw everything with a handmade pen: build point arrays, offset them by a smooth low-frequency wobble (a few summed sines, never per-point noise), stroke with round caps in one or two passes at slightly different width and opacity, scatter short scratchy hatch strokes, fill the heaviest masses solid black; then add a rules layer so nothing slides off the form — clamp parts inside the silhouette, keep anything worn or attached sitting on its outside, drop occluded parts on hard turns; keep the style deliberately naive and asymmetric — wobbly outlines, mismatched pairs, ink #232a22 on paper #e9e3d6, with occasional flat muted colour patches (terracotta, sage, dusty rose) laid under the ink. No libraries, just your Claude + JavaScript + Canvas 2D. And if you add a random seed to this, you get similar but always unique images.
Some of our favorite Claude Code projects we've seen lately: teapawt.net/mannay/status/20875220…
1
3
1,002
How Anthropic feel, releasing updates every day:
5
203
Anthropic published the official prompting guide for Fable 5. it's buried in the API docs and written for developers. here's how to use it for marketing work.👇
10
1
176
The short version: Fable 5 doesn't need a 500-word brief. It needs context, a clear output format, and defined stop points. Give it those three things and it runs marketing work end-to-end in a way that used to require an entire agency briefing cycle.
79
'/loop' for ongoing marketing tasks '/loop' lets Fable 5 work without manual intervention between steps. Useful for research-heavy work like competitive audits or content calendars. structure: /loop [time interval] + goal example: /loop 30 minutes, pull competitor positioning from these 5 URLs and flag any gaps vs our current messaging. to stop: /loop stop [loop name]
67
What to watch out for It runs longer than you expect on hard tasks. A full brand strategy at xhigh can take several minutes - that's normal. It can go beyond what you asked for. Fable 5 is proactive by design. Checkpoints from reply 5 fix this. Your old Claude prompts may actually produce worse results. Start fresh - a simple context-plus-request prompt beats an over-engineered saved prompt from Opus 4.8.
57
The full prompt structure for marketing work "I'm working on [brand strategy / competitive audit / budget plan] for [company and stage]. They need [what the output enables]. Request: [your specific ask in one sentence] Output format: [deck structure / spreadsheet / one-pager — exactly how you want it] Constraints: [assumptions it must not make — budget ceiling, channels off limits, audience restrictions]"
52
Build it a memory file for ongoing projects For brand strategy or long campaigns, Fable 5 performs better when it references previous decisions. Anthropic's recommended instruction: "Store one lesson per file with a one-line summary at the top. Record confirmed approaches and corrections alike. Don't duplicate what the chat history already has. Update existing notes rather than creating new ones. Delete anything that turned out to be wrong."
50
Set checkpoints for anything sensitive Fable 5 runs autonomously by default. For budget modeling or strategy work you'll want explicit stop points. Use this: "pause only when the work genuinely requires my input — a budget assumption you can't verify, a positioning call that changes direction, or something only I can confirm. otherwise keep going and report back when done."
48
Keep the prompt short Over-engineering kills output quality on Fable 5. You're constraining a model that would've figured out the right approach on its own. For a brand strategy brief: describe the company, the audience, the problem you're positioning against. Stop there. Let it build the structure.
46
Give it context, not just a task Fable 5 needs the why behind what you're asking. Vague briefs produce vague output. The formula: "I'm working on [brand strategy / campaign / budget plan] for [company type and stage]. they need [what the output enables — investor deck, Q3 launch, agency brief]. with that in mind: [your actual request]."
47
Match effort to the task low / medium: quick copy edits, simple research, one-liner rewrites high: default for most marketing work xhigh: full brand strategy, positioning overhaul, multi-channel campaign planning ultracode: autonomous end-to-end execution with minimal check-ins Don't run everything at xhigh. It's slower and more expensive. Save it for the work where quality is non-negotiable.
48
Fable 5 is not a better chatbot. It's built for autonomous, multi-step work - the kind that used to take a marketing team days of back and forth. Brand strategy, budget modeling, campaign planning, competitor research - these are exactly the tasks it's designed for. But only if you prompt it correctly.
1
115
⚡️@Anthropic dropped Claude Fable 5 today. Available now in the chatbot and Claude Code. free access runs until July 7 - after that, paid credits only. The benchmark numbers are hard to argue with. 80.3% on agentic coding (SWE-Bench Pro) vs 58.6% for GPT 5.5 and 54.2% for Gemini 3.1 Pro. 88% on Terminal-Bench 2.1. 78% on cybersecurity. Spatial reasoning gap vs GPT is especially wide - 38.6% vs 36.2% looks close until you see Gemini at 26.5%. Computer use is the one place Fable(AKA Mythos Lite) Preview still edges it out slightly - 85.4% vs 85.0%. The July 7 cutoff for free access is worth paying attention to if you've been sitting on testing it. What are you actually going to use it for first?
Fable 5 is back.
1
1
183
And you still understand that Mythos is way more powerful then Fable 5
Fable 5 is back.
1
2
109