AI Guides › Step-by-step guides

Screenshot Line to Aerial Camera Brief

By Nigel Guy · 6 min read

The usual mistake is to attach a map screenshot, type "make a drone shot of this" and expect the video model to understand the arrow you drew. It will not. The video models behind the connector take text and, optionally, a start image. Nothing in the setup accepts a path as a path. What you get is a plausible aerial of somewhere that looks a bit like your screenshot, with a camera move you did not choose.

The rule: Claude reads your line and turns it into words; Higgsfield only ever receives words and a start frame, so your job is to make those words specific and then check the result against the line.

Before you start

You need three things.

Credits are spent on every generation through the connector. Higgsfield states that the "unlimited" models and free generations on plans only work on higgsfield.ai, not through the connector. Its pricing page lists Kling 3.0 at 720p as roughly 7 credits per 5 seconds, and Veo 3.1 at roughly 40 credits per 4 seconds. Those figures move, so treat them as a guide to the gap between models, not a quote.

Step 1 — Add the Higgsfield connector

  1. In Claude on the web or in Claude Desktop, go to Customize > Connectors (Higgsfield's guide still calls this Settings > Connectors, so use whichever your menu shows).
  2. Click + Add, then Add custom connector.
  3. Name it Higgsfield and paste the server URL from Higgsfield's guide: https://mcp.higgsfield.ai/mcp. Click Continue.
  4. Accept the detected authentication settings unless you know you need something else, then click Add.
  5. Click Connect, sign in to Higgsfield in the window that opens and authorise.

Custom connectors are not verified by Anthropic, so only connect servers you trust. Claude's own guidance is to review permissions and tool approval requests, and to be wary of "Allow always".

Step 2 — Switch it on and test it

In a new conversation, use the + button at the lower left, then Connectors, and make sure Higgsfield is toggled on. Ask: "What is my Higgsfield credit balance?" If Claude answers without calling a tool, the connector is not active in that chat.

Step 3 — Set a spending rule

Higgsfield says there is no native credit cap. Start the conversation with: "Tell me the credit cost and wait for my yes before generating anything." This is a prompt instruction, not a hard limit, so keep an eye on the balance.

Step 4 — Draw the line and write the move

Draw your line, then write the move as an ordered list of what the camera does, using landmarks you can see: "rise from the street, follow the river, bank around the bridge, finish above the old town." Four beats is plenty for a short clip.

Step 5 — Upload the screenshot properly

Claude can look at your pasted screenshot to understand the line. But Higgsfield's help page says the agent cannot pass chat attachments to Higgsfield. A reference image has to go through the upload window the agent opens in the conversation. If Claude asks you to attach the file again, reply: "Open the Higgsfield upload window for my file." Remember that the image you upload carries your drawn line, so a model that treats it as a start frame may draw the line into the video. If you see that, upload a clean copy and keep the marked one for Claude only.

Step 6 — Send the master prompt

Fill in the brackets and paste this.

ROLE
You are my aerial-camera planner. You can see the screenshot I have attached and you have the Higgsfield connector available.

CONTEXT
The screenshot shows [PLACE_AND_WHAT_IS_VISIBLE]. I have drawn one line with an arrowhead on it. The tail is the start of the camera path and the arrowhead is the end.

GOAL
Produce one aerial clip of about [SECONDS] seconds that travels along the route I drew, one unbroken movement, with the look [LOOK: time of day, weather, lens feel].

STEPS
1. Describe my line back to me in plain words: start point, each bend (left or right), landmarks passed, end point. Stop and ask if anything is unclear or I have not filled in a placeholder.
2. Turn that into an ordered camera move: height, speed, direction changes and where the camera ends up looking. My notes on the move: [MY_MOVE_NOTES].
3. Recommend one video model from [MODELS_I_AM_HAPPY_WITH] and say why, with the credit cost per clip.
4. Wait for my approval of the move, the model and the cost.
5. Upload the screenshot through the Higgsfield upload window if the model can use a start frame, then generate.

CONSTRAINTS
- Natural drone speed, no fisheye, no warping of buildings or roads.
- Do not invent landmarks that are not in the screenshot.
- Spend no more than [MAX_CREDITS] credits in total.

OUTPUT
After generating, report: model used, credits spent, and a three-line check of the result against my line (start, bends, end), marking each as matched, partly matched or missed.

SELF-CHECK
Before generating, confirm that the move description contains every bend I drew, in order.

Check it worked

Compare the clip against the line, not against how pretty it is. Does it start where the tail is? Does it bend the way you drew, in the same order? Does it end near the arrowhead? Ask Claude for the matched, partly matched or missed report and then judge it yourself. Expect "partly matched" often. Higgsfield's connector gives no control over camera path beyond what the text and any start frame can say, and I could not find documentation promising that a drawn line will be followed, so treat any match as a bonus. Generations also appear in your Assets on higgsfield.ai, usually within about 60 seconds.

What to skip

Guardrails

Sources

All 751 AI guides · JulieMango plans from £17/mo